Democracy vs.

dictatorship: comparing the

evolution of economic growth under two
political regimes
Marc Schiffbauer1 and Ling Shen2

Abstract

A democratic society is often regarded as a prerequisite for economic growth and development.
Yet, most empirical studies are not able to identify a positive link between GDP-growth and
democracy indexes. In addition, it is an empirical stylized fact that (i) most developing countries
are dictatorships and (ii) many poor dictatorships experienced high growth performances and
emerged from poverty such as South Korea, China, Egypt, etc. Against this background, it is of
interest to analyze in which way the growth performance between autocratic and democratic
economies may differ, in particular among low income countries. To answer this question, we
compare the endogenous growth paths of two economies that exclusively differ in their political
regimes in the context of an overlapping-generations model. In particular, the key features of the
model are (i) a positive bequest motive in the form of investments in education or productive
public capital (infrastructure), (ii) a higher marginal (inter-temporal) utility of consumption today
versus consumption tomorrow in low-income countries (e.g. subsistence level of consumption),
and (iii) a dictator that cares about her income or the income of her dynasty tomorrow (compare
McGuire and Olson (1996)). In this framework, we demonstrate that poor but stable dictatorships
exhibit a higher equilibrium growth rate than comparable (equally poor) democracies. Moreover,
there exists a particular threshold value in income such that the growth-reducing impact of
dictatorial consumption (corruption) outweighs the higher (initial) public investments. Thereafter,
the growth rate under democracy dominates the one in dictatorship.

Keywords: Dictatorship, Democracy, Political transition, Economic Growth

JEL Classification Numbers: D72 D74 O12 P16

Ling Shen wants to thank “Shanghai Pujiang” project (2007) for their generous financial support.
This project is also supported by Leading Academic Discipline Program, “211” Project for
Shanghai University of Finance and Economics (the 3rd phase).

1
Correspondence: Bonn Graduate School of Economics, Bonn University, Bonn 53175, Germany. Email:
[email protected]
2
Correspondence: School of Economics, Shanghai University of Finance and Economics, Guoding Road 777,
200433, Shanghai, China. Email: [email protected]

1
1 Introduction

Does democracy spur growth? Or to put it in other words: does political freedom
involve economic freedom? In fact, a democratic society is often regarded as a
prerequisite for economic growth and development. Yet, most empirical studies are
not able to identify a positive link between GDP-growth and democracy indexes, e.g.
Barro (1996), Tavares and Wacziarg (2001). Furthermore, it is an empirical stylized
fact that (i) most developing countries are dictatorships while most developed
countries are democratic and (ii) some poor dictatorships experienced high growth
performances and emerged from poverty such as Vietnam, Egypt or China and South
Korea, Taiwan, Mexico, or Ecuador before their democratization (compare Table 1).
Against this background, it is of interest to analyze under which conditions a
democracy outperforms a dictatorship in terms of higher economic growth? Therefore,
we contrast the endogenous growth paths of two economies that exclusively differ in
their political regimes in the context of an overlapping-generations model.

As per the relationship between political institutions and economic growth, the
empirical literature mainly relies on three institutional indices: (i) the International
Country Risk guide (e.g. Knack and Keefer (1995) or Hall and Jones (1999)), (ii) an
aggregate index of survey assessments of government effectiveness from the World
Bank (Kaufmann et al. (2003)) and (iii) the Polity data set collected by Jaggers and
Marshall (2000). However, all of these measures are endogenous with respect to
growth and development. Glaeser et al. (2004) document that in particular the first
two indices measure political outcomes instead of constraints. That is, they are highly
volatile3 and in fact uncorrelated with direct measures of political constraints on
government coming from either electoral rules or courts. Moreover, the authors reveal
that the indicators are constructed in a way such that dictators choosing good policies
also receive high evaluations. To address the problem of endogeneity Acemoglu et al.
(2001) exploit historical information on e.g. settler mortality in the 18th and 19th
century to instrument for institutions. The authors detect a positive correlation
between the historical instruments and the International Country Risk guide or current
levels of GDP. Yet, their work is criticized by Engerman and Sokoloff (2003) or
Glaeser et al. (2004) who show that their historical institutional instruments are also
strongly correlated with alternative growth determinants - e.g. schooling. In sum, the
relevant empirical growth literature is based on political outcomes instead of
constraints. This observation is important because good political outcomes appear to
be consistent with autocratic regimes in some countries. Glaeser et al. (2004) (p.286)
conclude: “(...) it is crucial to understand what makes a successful dictatorship.” In
the following, we analyze under which circumstances dictatorships are successful and
potentially outperform comparable democratic regimes in terms of economic growth.

The model we put forward is consistent with the indeterminant empirical results on
democracy (political constraints) and growth, because both democratic and autocratic

3
The indices are highly volatile and hence can not represent permanent deep institutions.

2
societies create distortions. Successful economic performances will come from
democracies and dictatorships that feature relatively high investments in future
generations, represented by investments in education and infrastructure, relative to
current consumption (redistribution). In the economy developed below it turns out
that relatively rich democracies and relatively poor but stable dictatorships choose
less distortional public policies. In contrast, poor but unstable dictatorships generate
the worst outcome. In this regard, we show that poor dictatorships are more likely to
be stable if (i) the economy is large and (ii) the dictator has a higher survival
probability or lower enforcement costs, e.g. due to a lower degree of ethnic diversity
in the economy.

The model is based on three key ingredients. First, there exists a positive bequest
motive in the form of investments in education or productive public capital
(infrastructure). This assumption involves a tradeoff between consumption and
productive investments in future generations and is standard in theories of economic
development, compare e.g. Aghion and Bolton (1997). Second, we suppose a higher
marginal (inter-temporal) utility of current consumption versus consumption of the
next generation in low-income countries, for example due to a subsistence level of
consumption. This reasoning follows e.g. Galor and Weil (2000) or Greenwood and
Jovanovic (1990) and implies that the optimal inter-generational consumption
decision may be constrained in poor economies.4 Third, we assume, in accordance
with McGuire and Olson (1996) and Shen (2007), that a dictator cares about her own
consumption and the income of his direct offspring. 5 In this framework, we
demonstrate that poor democracies exhibit a lower equilibrium balanced growth rate
than equally poor but stable dictatorships since the median voter under-invests in
productive capital. In contrast, under specific conditions which we outline below, the
dictator of a poor economy invests in infrastructure capital to spur economic growth
and hence the income of her offspring tomorrow.

Intuitively, the low income of the median voter in a poor democracy involves a
relatively low marginal inter-generational utility which constraints productive
investments in future generations. On the other hand, the higher income level (from
expropriation) of a dictator, in an otherwise identical economy, implies a higher
marginal inter-generational utility of the decision-maker and hence enhanced public
investments. If the economy is relatively poor, this effect outweighs the
growth-retarding impact of corruption in dictatorship (dictator’s consumption). Thus,
the dictator of a less developed country may invest in infrastructure while the median
voter calls for redistributive policies to achieve a higher current consumption share.
This redistribution versus public investments policy tradeoff in democratic regimes
has been informally stressed by Barro (1996) and others.6 In addition, our model

4
Steger(2000) analyzes the performance of growth models with subsistence consumption with respect to
empirical stylized facts of economic development.
5
McGuire and Olson (1996) and Shen (2007) examine the incentives of a dictator to spur economic growth.
6
Barro (1996) (p.1): “These (growth retarding) features (of democracy) involve the tendency to enact
rich-to-poor-redistribution of income in systems of majority voting and the enhanced role of interest groups in

3
involves a particular threshold value in income such that the growth-reducing impact
of dictatorial consumption (corruption) outweighs the higher (initial) infrastructure
investment. Thereafter, the growth rate under democracy strictly dominates the one in
dictatorship.

Barro (1996) employs cross-country growth regressions to investigate the democracy

growth nexus. He reports a weakly negative effect of democracy on economic growth
once the effect of rule of law, free markets, human capital and initial GDP are
controlled for. Acemoglu et al. (2005) argue that there is no causal effect of income on
democracy after controlling the country fixed effects. Tavares and Wacziarg (2001)
examine the importance of different potential transmission channels for the effect of
democracy on growth applying panel data for 65 countries.7 The former is measured
by the Freedom House indicator, which is based on information on political
procedures instead of politic outcomes. More specifically, the authors estimate the
impact of democracy on different determinants of economic growth via a system of
simultaneous equations. They find that democracy fosters growth by enhancing
educational attainments while it hinders growth by reducing the rate of physical
capital accumulation. These results are consistent with the specific tradeoff
highlighted by our model: the median voter (in relatively poor countries) forgoes
public capital investments in favor of redistributive policies.8

The theoretical part of the literature analyzes both directions of causality between
democracy and growth. Acemoglu and Robinson (2000) develop a model of political
transition: economic growth initially leads to higher income inequality which in turn
puts pressure on the autocratic elites to introduce democracy in order to prevent social
unrest. Bourguignon et al. (2000) assume that education not only affects economic
growth but also political participation. Cervellati et al. (2005) consider an endogenous
evolution of economic and political institutions in the form of a state of law. McGuire
and Olson (1996) and Shen (2007), on the other hand, characterize initial conditions
in a way that it is optimal for an income-maximizing dictator to introduce
growth-promoting policies. In particular, they establish that initial capital stocks of the
citizens must not fall below a certain threshold. To the best of our knowledge, the only
additional study that compares economic performances of different political regimes
within a unified endogenous growth framework represents the work of Acemoglu
(2003). However, his study focuses on the choice of property rights by oligarchic and
democratic regimes instead of a tradeoff between redistribution and public
investments in relatively less developed democracies.

In the succeeding section, we outline the model and develop the theoretical results.

systems with representative legislatures. ”

7
The authors report an insignificant bivariate correlation between democracy and growth. Hence, they conclude that
positive and negative growth effects from democracy cancel each other out in the sample.
8
Our study is also related to the literature on income inequality and economic growth, e.g., Barro (2000). In contrast
to Barro, however, the opportunity to expropriate future income provides incentives for a dictator to invest in
infrastructure.

4
Section 3 lists some empirical stylized facts concerning the economic performance of
the different political regimes. The final section concludes.

2 The model

In the following, we outline the dynamic overlapping-generations model. First, we

illustrate the democratic economy, thereafter we introduce dictatorship. The model
economy is identical for both political regimes apart from (i) the determination of
public policies consisting of a tax rule and the public provision of infrastructure
capital, (ii) a survival probability for the dictator, and (iii) differences in enforcement
costs between both political regimes. We define, for convenience, that each individual
runs her own firm to produce a homogeneous final good ( yi , t ). A separation between

entrepreneurs and households would yield identical results as long as goods and labor
markets are competitive and production is deterministic. Moreover, we abstract from
modeling the dynamics of a heterogeneous income distribution in order to keep the
comparison between both political regimes tractable. This is less restrictive than it
might first appear. First, the dictator’s optimization problem can supposed to be
independent of the distribution of income since her ‘tax base’ exclusively depends on
the aggregate income level (compare section 2.2). Second, in the case of democracy,
we need to impose an additional assumption on the initial income distribution, which
we outline in Remark 1 below, to generalize our main result accordingly.

2.1 Democracy

In each period t a generation consists of a mass N of identical, risk averse

individuals. We abstract from population growth in that we assume that each
individual i has a single parent so that N is constant over time. Members of
generation t live for two periods. In the first period of life (childhood), t − 1 , the
individual exclusively consumes education. In the second period, she produces a
single homogeneous final good ( yi ,t ), receives the payoffs from her production,

consumes, pays for the education of her child ( bi ,t ) and votes for public policy rules.

Individual production occurs according to a constant return to scale technology:

yi ,t = Ahiα, t Gt1−α
(1)

where A is a constant technology parameter, hit the individual’s human capital

stock and Gt the public provision of non-excludable, non-rival infrastructure

5
capital. 9 The final good can be used for consumption or bequest. The bequest
comprises investments in the education of the child ( bi ,t ) or investments in public

infrastructure capital ( I t ) since we assume that the latter is not productive until the

following period. 10 The stock of human and infrastructure capital follow the
difference equations:
hit = hit −1 + bit −1 , hi 0 > 0 (2)

Gt = Gt −1 + I t −1 , G0 > 0 (3)

where hi 0 > 0 implies that individuals are productive even in the absence of formal

education.11

The preferences of members of generation t are defined over consumption as well as

the potential individual income of their children. 12 However, we assume that a
subsistence consumption constraint is binding for a sufficiently low level of income.
~
In particular, we define the subsistence level of consumption ( c ) as the level that
induces an individual to consume all of her (net) income instead of leaving a positive
bequest for her children, e.g. by investing in the education of her children. We follow
Galor and Weil (2000) in modeling the subsistence consumption level.13 That is, we
assume that the utility is strictly monotonically increasing and strictly quasi-concave
for a sufficiently high income level, satisfying the conventional boundary conditions
that ensure the existence of an interior solution for the utility maximization problem.
Yet, if the income level is sufficiently low, the consumption constraint is binding
involving a corner solution with respect to the consumption/bequest decision. In this
case, the income level is so low that an individual is not able or willing to save part of
her income for investments in her offspring’s education. Accordingly, she spends all
of her net income for basic needs like food, shelter, or clothing. These preferences are
defined by the following log-linear utility function:14

9
We do not introduce an individual capital stock explicitly, but hi can be thought of as a composite of private
and human capital as long as capital/financial markets are complete.
10
This assumption is justified, for example, if the construction of infrastructure capital persists one time period.
We note that our qualitative results do not hinge on the time-to-build assumption of infrastructure capital. Yet, this
assumption simplifies the analysis.
11
We introduce hi 0 > 0 to rule out corner solutions of zero production.
12
This component of the utility function may represent either intergenerational altruism or implicit concern about
potential support from children in old age (see Galor and Weil (2000).
13
Alternatively, a qualitatively similar result can be derived by assuming a Stone-Geary utility function of the
form U = β ln(cit − c ~) + (1 − β ) ln( y ) . However, the adoption of this formulation would increases the
it +1
dimensionality of our model.
14
This assumption is commonly used in the development literature in order to model a higher marginal utility
from contemporaneous versus future consumption for low levels of income which is typically found in the data. In
addition to Galor and Weil (2000), who apply an analogous definition of the subsistence consumption level, we
show in the appendix that our utility function can also be derived from the one of Galor and Moav (2006).

6
 ⎧0 < β < 1 if cit ≥ c~
U = β ln(cit ) + (1 − β ) ln( y it +1 ), ⎨ (4)
⎩β = 1 if cit < c~

where preferences over current consumption and future income are assumed to be
additively separable and exogenously weighted by β .15 We show in the appendix that

the utility function in (4) can be derived directly from the utility function of Galor and Moav

(2006) who assume the following preferences: U = β ln(cit ) + (1 − β ) ln( y it + θ ) , where θ

denotes the subsistence consumption level. Furthermore, the individual budget constraint is
given by
bit = (1 − τ t ) yit − cit (5)

where τ t is a tax rate.

The tax rate and the public investment in infrastructure are determined by the
government.16 Note that a redistributive policy, in the form of contemporaneous
income transfers between individuals, is redundant since we assume that the income is
equally distributed across the individuals.17 Consequently, redistribution at the cost of
public infrastructure investment is equivalent to a lower tax rate in this setting. That is,
we reduce the three dimensional tradeoff between higher taxes, redistributive policies
and infrastructure investments to a two dimensional tradeoff between taxes and
infrastructure provision. The assumption of a homogenous income distribution
N
simplifies the aggregation problem of individual production to Yt = ∫0 yit di = Nyit .
Hence, the government budget constraint amounts to

I t = τ tYt = τ t Nyit (6)

In a democratic society, τ is determined by majority voting which characterizes the

democratic regime.

The maximization problem for an individual in the case of democracy is given as

15
The dependence on income of next generation instead of consumption means that parents care about the income
of their children, but not the way it is spend (e.g. split between consumption and bequest to the second next
generation). We note that U = β ln(cit ) + (1 − β ) ln(cit +1 ) would yield analogous results.
16
Infrastructure capital can not be provided privately in this setting due to a free-rider problem stemming from the
non-excludability of G.
17
However, such effects are potentially important since redistributive policies in the democratic regime influence
the income of the majority voter. We discuss the robustness of our results with respect to an endogenous income
distribution in Remark 1. Yet, we already note that an average equilibrium income level below the subsistence
consumption level is sufficient to ensure a poverty trap in democracy even in the case of a right-skewed
endogenous income distribution since the equilibrium income of the median voter can at most comply with the
average individual income level.

7
follows:
max
τ
U ,
cit ,
i s.t. (1), (2), (3), (5), (6)
t

whereby the interior solutions for the first-order conditions amount to:

∂U i β (1 − β ) αyit +1
: = (7)
∂cit cit yit +1 hit +1

∂U i
: 0=
(1 − β ) ⎡αyit +1 (− y ) + (1 − α )yit +1 Ny ⎤
⎢ it ⎥ (8)
∂τ t
it
yit +1 ⎣ hit +1 Gt +1 ⎦

We solve (7) for the optimal (unconstrained) level of consumption, which is a positive
function of net income:

β [hit + (1 − τ t ) yit ]
cit∗ = (9)
β + (1 − β )α

The individual budget constraint, the utility function (4) and the solution for cit∗
define the optimal rule for investments in education of the next generation:

⎧ ⎫
⎪0 if (1 − τ t ) yit ≤ c~ ⎪
⎪⎪ ⎪⎪
bit = ⎨(1 − τ t ) yit − c~ if cit∗ < c~ < (1 − τ t ) yit ⎬ (10)
⎪α (1 − β )(1 − τ ) y − β h ⎪
⎪ t it it
if cit∗ ≥ c~ ⎪
⎩⎪ α + β − αβ ⎭⎪

It follows from the non-linearity in (4) that we have to distinguish three different
cases for an individual’s bequest decision depending on her net income. First, she
consumes all of her net income if the latter is below the subsistence level. In the
second case, net income exceeds the subsistence consumption level, but the optimal
unconstrained consumption level given in (9) falls below it. In this case, an individual
invests a constraint optimal, but positive part of her net income in the education of her
offspring. Third, each individual invests the optimal amount of her net income in the
education of her offspring if the optimal consumption level exceeds the subsistence
consumption level. The optimal taxation rule can be computed from (8). Since all
individuals are the same the individual tax decision complies with the one of the
median voter:
1−α
N (hit + bit ) − Gt
τ t∗ = α (11)
Nyit

8
The optimal tax rate is a positive function of the investment in human capital ( bit )

which is in turn determined by the three different investment rules defined in (10).
Thus, we have to distinguish between three different cases depending on the
individual contemporaneous income level. In the following, we derive the threshold
levels in income which determine the corresponding investments in human and
infrastructure capital.

Case 1: yit ≤ c~ :

In the first case, where (1 − τ t ) yit ≤ c~ , the individual consumes all of her net income

so that bit = 0 . Hence, the aggregate human capital stock remains constant. Moreover,

we infer from (4) that the median voter chooses zero infrastructure investments
( τ = 0 ) so that the above condition simplifies to y ≤ c~ . In this case, the
t it

predetermined levels of infrastructure and aggregate human capital are equal to their
initial values: Gt = G0 and Nhit = Nhi 0 . Intuitively, the individual income level is so

low that individuals maximize their utility by spending all of their income for
subsistence consumption needs involving zero investments (bequest) in future
generations. It follows that the democratic society suffers from the absence of
long-run economic growth. The urgent consumption needs today prevent individuals
from growth promoting investments in future generations.

Proposition I: Given the preferences and production specifications defined in (1)-(5),

it follows that a democratic economy, characterized by majority voting, is kept in a
long-run zero-growth trap if the initial income of the median voter falls below the
subsistence consumption level ( y ≤ c~ ).
i0

Proof : The proof follows directly from (1), (4), (10) and (11) since bit = 0 and

τ t = 0 if yit ≤ c~ .

We demonstrate in the appendix that the implementation of the alternative individual

preferences defined in Galor and Moav (2006) lead to equivalent results in our model.
In particular, Galor and Moav (2006) assume the analogous preferences:
U = β ln(cit ) + (1 − β ) ln( yit + θ ) , where θ denotes the subsistence consumption level.

Their preference specification yields an explicit threshold for the subsistent income
level; however, the adoption increases the dimensionality of the model. As we are not
interested in the explicit value of subsistent income as a function of θ , we do not

9
implement these alternative preferences.

Moreover, note that the specifications in (1)-(5) involve the existence of an individual
bequest motive and an exogenous income distribution. However, the latter assumption
is less restrictive than one might guess since redistributive policies in the case of a
heterogeneous symmetric or left-skewed income distribution would not alter the
median voter’s equilibrium income level (the average individual income is below the
median).18 Moreover, the above condition can be extended to ensure the existence of
a democratic poverty trap in the case of a right-skewed distribution. That is, one need
to assume that the average equilibrium income level of an individual is below the
subsistence consumption level since the income level of the median voter can at most
comply with the average individual income after redistribution.

Remark 1: We need to assume additionally that the initial average equilibrium income
level is below the subsistence consumption level ( y ≤ c~ ) to generalize Proposition 1
i0

to the case of an endogenous (heterogeneous) income distribution. To understand this

condition, note that some individuals, whose income levels are above the subsistence
consumption level, invest in education in this case. This raises aggregate human
capital and income next period. Yet, diminishing returns in individual human capital
investments eventually lead to constant aggregate human capital and income levels in
equilibrium. The income distribution that corresponds to the equilibrium income level
is constant if it is symmetric or left-skewed. If it is right-skewed, the median voter
votes for redistributive policies to augment her income. However, the income level of
the median voter after redistribution can never exceed the average income level. Thus,
in the case of an endogenous (right-skewed) income distribution, it is sufficient to
assume that the initial average equilibrium income level is below the subsistence
consumption level in order to ensure the existence of a democratic poverty trap
analogue to Proposition I.

Case 2: yit > c~ > cit∗

This case includes two sub-cases. First, it is optimal for an individual to choose
cit = c~ , i.e., bit = (1 − τ t ) yit − c~ > 0 , if cit∗ < c~ < (1 − τ t ) yit . Second, the individual

chooses cit = (1 − τ t ) yit , i.e., bit = 0 if (1 − τ t ) yit ≤ c~ . The distinction of cases

depends on the ratio of the initial infrastructure capital stock to the aggregate human
capital stock.

18
The equilibrium income level refers to the level that is installed after the aggregate human capital stock adjusted
~ would
to its equilibrium level relative to the infrastructure capital stock. That is, some individuals with y i 0 > c
initially invest in education. Yet, an income of the median voter below the subsistence consumption level ensures
the absence of infrastructure investments which would eventually prevent further individual human capital
investments.

10
If the predetermined aggregate human capital stock is abundant relative to the
infrastructure capital stock, the median voter chooses a positive tax rate even if
educational investment is zero. From (11) it implies:

1−α
Nhit − Gt
1−α
τ t (bit = 0) = α > 0 ⇔ Gt < Nhit (13)
Nyit α

This initial increase in infrastructure capital causes a temporary rise in individual

income. If the new equilibrium income level does not exceed the above condition,
1−α
both stocks will be balanced: Gt = Nhit . Not surprisingly, the optimal ratio of
α
the two input factors depends on the shares of both input factors in the individual
production functions.

1−α
On the other hand, if the initial capital stocks satisfy Gt > Nhit , then the optimal
α
tax rate equals zero. In this intermediate case, there are no contemporaneous
infrastructure investments. Still, individual income grows over time as investments in
human capital are positive. In this sub-case, bit declines over time because of τ t = 0

due to diminishing returns, but remains always positive since by definition bit is

strictly positive if yit ≥ c~ > cit∗ holds. That is, the economy will eventually satisfy
1−α
condition Gt = Nhit . Thus, we can infer from both sub-cases that the increase in
α
1−α
human capital (or infrastructure) will cause condition Gt = Nhit to be satisfied
α
over time so that eventually the growth rate increases until it reaches the
unconstrained growth path defined below (Case 3). Consequently, there exists no
balanced growth equilibrium as long as yit ≥ c~ > cit∗ (Case 2). Economic growth is
restricted initially, but will always converge to the unconstrained growth path in this
intermediate case. Yet, the economy may experience different growth performances
during this transitional phase: (i) the growth rate declines over time if the ratio of the
initial human capital stock relative to the infrastructure stock is relatively unbalanced
(e.g. condition (13) holds), (ii) it accelerates otherwise since the restriction on future
investments ( b , τ ) alleviates since y is growing while c~ is constant. In other
it t it

words, democracies that feature an initial intermediate aggregate income level, such
that individual consumption of the median voter is above the subsistence, but below
the unconstrained level, experience a relatively low, but strictly positive GDP-growth

11
rate in the beginning, which may initially decrease or increase, but will eventually
accelerate at some point in the future until it reaches the long-run unconstrained
balanced growth path.

Case 3: cit∗ > c~

If the optimal consumption level is above the subsistence level ( cit∗ > c~ ), the optimal
tax and bequest decisions amount to:

τ t* =
(1 − α )(1 − β )N (hit + yit ) − (α + β − αβ )Gt (14)
Nyit

⎡G ⎤
bit = α (1 − β )⎢ t + hit + yit ⎥ − hit
∗
(15)
⎣N ⎦
Investments in infrastructure and aggregate human capital are balanced if
1−α Gt 1 − α
Gt = Nhit . Henceforth, we define Z dem = = N . If the predetermined
α hit α
input levels are unbalanced, more resources are invested in the scarce input factor
until the endowments of both levels are balanced. Thus, in the long-run the economy
sustains positive endogenous balanced GDP-growth and the ratio of infrastructure to

individual human capital is characterized by Z dem . In particular, the balanced growth

rate amounts to:

g dem
=
Nyit +1 ⎛
⎜
−1 = φ A +
(
Z dem
α
) (
+ Z dem
α −1 ⎞
)
⎟ −1
Nyit ⎜ N ⎟
⎝ ⎠ (16)
α⎡ α
⎛ α ⎞ 1 ⎛1−α ⎞ ⎤
( )(
= 1−α 1− β ⎜ ) ⎟ ⎢ AN
(1−α )
+ ⎜ ⎟ ⎥ −1 ≡ g
dem
>0
⎝ 1 − α ⎠ ⎢⎣ 1 − α ⎝ α ⎠ ⎥⎦

Gt 1 − α
where φ = [α (1 − β )]α [N (1 − α )(1 − β )](1−α ) and Z dem = = N.
hit α

2.2 Dictatorship

In the case of dictatorship, one of the otherwise identical individuals determines the
expropriation (tax) rate and the public provision of infrastructure capital. Hence, this
agent, which we refer to as the dictator, features the same preference as the other
individuals. 19 She maximizes her utility function, which depends on her current

19
The variables referring to the dictator are marked with the subscript r.

12
consumption and the income of the next generation - her direct offspring. Yet, there
are two key differences with respect to other households in that she (i) is potentially
richer due to the opportunity to expropriate private income and (ii) could have no
offspring because of a revolution. The introduction of such a dictator potentially
curtails the utility of all other individuals. In particular, a necessary condition for a
reduction in the contemporaneous welfare of individuals due to dictatorship in our
framework is a higher tax (expropriation) rate relative to democracy. This condition is
certainly satisfied if the subsistent consumption constraint is binding: the tax rate in
set to zero in democracy while it is always strictly positive in dictatorship.
Nevertheless, economic growth can be positive in dictatorship as opposed to
democracy if the ruler invests a positive fraction of the expropriated income in the
economy to increase the income (expropriation base) of her offspring. In the
following, we derive the necessary condition for this scenario to come true.20

The dictator simultaneously determines the tax rate, her consumption level, and her
bequest, in the form of investments in infrastructure, in order to maximize her utility
function:

⎧⎪0 < β < 1 if ctr ≥ c~

max U r
= β ln(c tr ) + (1 − β )ψ ln( y tr+1 ), ⎨ (17)
ctr ,τ t ⎪⎩β = 1 if c tr < c~

where ψ is the probability of survival of dictatorship, subject to (1), (2), (3), (5), the

budget constraint of the dictator

y tr = ϑτ t Ny it = c tr + btr (18)

and the citizen’s utility function ( U i ):

⎧0 < β < 1 if cit ≥ c~ ⎫

U i = β ln cit + (1 − β )ln((1 − τ t +1 ) yit +1 ) ⎨ ⎬ (19)
⎩β = 1 if cit < c~ ⎭

The parameter 0 < ϑ < 1 in the dictator’s budget constraint represents the enforcement
costs of expropriation in dictatorship. We suppose that dictators have to spend greater
resources than democratic authorities to enforce their tax revenues ( ϑ < 1 ) which reduce their
tax revenues. Moreover, the citizen’s utility function is amended to account for the fact that
part of the individual future income is expropriated by the future dictator and hence
partly lost to dictatorial consumption (corruption). The corresponding first-order
conditions amount to:

20
We emphasize that our model is not appropriate for a comparative welfare analysis between dictatorship and
democracy since we do not account for, e.g., the utility from unrestricted property rights or political freedom. Instead,
we exclusively focus on the comparison of the evolution of economic growth under both political regimes.

13
 ∂U r
= 0:
β
=
(1 − α )(1 − β )ψ (20)
∂ctr ctr Gt +1

∂U r (1 − β )ψ
=
⎡ αy ∂b
ϑτ t +1 N ⎢ it +1 it +
(1 − α ) yit +1 ϑNy ⎤
it ⎥ (21)
∂τ t r
yt +1 ⎣ hit +1 ∂τ t Gt +1 ⎦

Condition (20) determines the optimal level of infrastructure investments by the

dictator:

btr =
(1 − α )(1 − β )ψytr − βGt (22)
(1 − α )(1 − β )ψ + β

Note that btr is independent of bit since the dictator does not care about
contemporaneous consumption decisions of her citizens.21 In fact, the dictator always
invests in infrastructure ( btr > 0 ) if:

β Gt
y it > ≡~
good
y (23)
(1 − α )(1 − β )ψϑτ t N

This condition defines “when the dictator will be good”. In this case, it is optimal for
the (inter-temporal) income maximizing dictator to invest in infrastructure if the
individual income of citizens exceeds this threshold value. We can infer from (23) that
the probability of growth-promoting infrastructure investments by the dictator
increases in the size of the population while it decreases in the predetermined (initial)
level of the infrastructure capital stock. Moreover, the probability of a ‘good dictator’
is increasing in ψ and ϑ . Thus, a decline in the survival probability of the dictator

raises her incentives for corruption. This result is consistent with the analogous
findings of McGuire and Olson (1996) or Shen (2007). In particular, dictators become
“roving bandits” if ψ → 0 . Similarly, higher enforcement cost (an decline in ϑ )

leads to higher corruption and hence lower infrastructure investments in dictatorship.

Following, Easterley and Levine (1997), we expect that ψ and ϑ are decreasing in
the degree of a country’s ethnic diversity. That is, the higher a country’s degree of
ethnic diversity, the lower is a dictator’s probability of survival and the higher is her
enforcement cost due a more intense contest for political power among different
ethnic groups.
21
Yet, the investment decision of the dictator is a positive function of the predetermined aggregate stock of
human capital and hence of past individual investment decisions.

14
Condition (21) demonstrates that the change in the marginal utility of the dictator due
∂U r
to marginal changes in the expropriation rate ( ) depends on the optimal bequest
∂τ t

∂bit ∂bit
behavior of the citizens ( ). depends, in turn, on the stage of development
∂τ t ∂τ t

of the economy (compare (10)). Hence, we need to consider the three different cases
∂U r
defined in (10) in order to compute .
∂τ t

Case 1: yit ≤ c~ :

∂bit
In the first case, we know from (10) that bit = 0 . It follows that = 0 and
∂τ t

∂U r
hence > 0 . Thus, the marginal utility of the dictator is strictly increasing in the
∂τ t
expropriation rate. An interior solution for τ does not exist in this case. It follows
that the dictator expropriates the highest possible value. However, we suppose that the
dictator can not expropriate all of the individuals income ( τ < 1 ) due to the existence
of backyard production (black market) which can not be taxed.22 It follows that the
optimal expropriation rate equals to: 0 < τ t = τ < 1 .

We assume that the dictator can expropriate her citizens even if their consumption
level is below the subsistence one. In this regard, it is important to keep the definition
of the subsistence consumption level in mind: if c < c~ an individual rather spends
it

all of her net income for basic needs like food, shelter, or clothing than investing in
the education of her children.
Consequently, the dictator always invests in infrastructure, even if the economy is
“relatively poor”, as long as condition (23), which determines the existence of a ‘good
dictator’, is satisfied. A comparison of the two individual income levels which define
subsistent consumption ( y ≤ c~ ) and the existence of a good dictator ( y ≥ ~
good
it y
it )

reveals the characteristics of an economy in which both income levels overlap. In

particular, the threshold level in individual income that ensures positive economic
growth in dictatorship is strictly lower than the corresponding threshold level in

22
Compare Acemoglu and Robinson (2000).

15
democracy if N and/or ψ is large:

~ βGt βGt
y good < c~ ⇒ N > or ψ >
(1 − α )(1 − β )ψϑτ c~ (1 − α )(1 − β )Nϑτ c~
(24)
βGt
⇒1>τ > >0
(1 − α )(1 − β )ψϑNc~

Thus, a poor democracy that is kept in a zero-growth trap (Case 1) would sustain
positive endogenous growth in dictatorship if (i) the size of the population ( N ) is
large relative to the predetermined (initial) infrastructure capital stock, and/or (ii) the
dictatorial regime would be relatively stable (implied by a greater ψ ). The

introduction of a dictator would lead to positive growth in this case. The economy
overcomes the zero-growth trap even though the utility of individuals initially
declines. Note, however, that the dictator would become a ‘roving bandit’ if condition
(24) is violated either because her survival probability would be too low or the size of
the population too small.

Hence, we obtain a range of individual income levels, which are below the subsistence
level, such that the democracy features zero-growth in the long-run while a “good
dictatorship”, defined in (23), exhibits positive economic growth. Moreover, we know
1−α
from the conditions Gt > Nhit and τ t = τ < 1 that individuals in the autocratic
α
regime will invest in human capital of the next generation bit > 0 ( yit > c~ ) before
diminishing returns in infrastructure capital constrain further investments. This result
is summarized in the following proposition.

Proposition II: Given the preferences and production specifications defined in

(1)-(5), (17)-(19), it follows that the long-run equilibrium growth rate of a relatively
~ good < y < c~
poor dictatorship ( y it ) exceeds the one in an equally poor democracy if
~ good
the condition for a “good dictatorship” ( y it > y ), which is defined in (23), is
satisfied. That is, the threshold level in individual income that ensures positive
long-run economic growth in dictatorship is strictly lower than the corresponding
~ good < y < c~
threshold level in democracy ( y it ) if, according to (24), the size of the
population is relatively large and/or the dictatorial regime is relatively stable.

It follows that a country can experience strictly positive economic growth in

dictatorship even though it would be locked in a zero-growth trap in democracy.
Intuitively, the result follows from two features of the model: (i) the dictator obtains a

16
positive utility from the income of her offspring and (ii) her income is always above
the subsistence level. Hence, she might always invest in the economy (infrastructure)
if it is large enough and stable in order to expropriate more (aggregate) income in the
next period. The positive infrastructure investment generates transitional growth and
hence a higher marginal utility from individual human capital investments. This
finally generates private investments in human capital ( bit > 0 ) and hence positive
endogenous growth. The result is consistent with the relatively good past performance
of some poor autocratic economies - compare section 3.

Thus, a relatively poor dictatorship, which satisfies condition (23), initially

experiences slow but stable economic growth until individual net income exceeds the
subsistence consumption level c~ (Case 2, 3).

Case 2: yit ≥ c~ > cit*

Again, we have to distinguish two sub-cases. First, the individual chooses
cit = (1 − τ t ) yit and bit = 0 if (1 − τ t ) yit ≤ c~ . In this case, the dictator imposes τ t = τ

∂U r
since her marginal utility from a marginal tax increase is strictly positive ( > 0 ).
∂τ t
This implies a positive growth rate due to public infrastructure investments. The
temporary rise in individual incomes eventually leads to the second sub-case:
cit∗ < c~ < (1 − τ t ) yit which involves cit = c~ and bit = (1 − τ t ) yit − c~ > 0 . In this case,
it follows from condition (10) that:

⎧ Gt +1 (1 − α )N ⎫
⎪> 0 if < ϑ ⇔ Z dic < Z demϑ ⎪
⎪ hit +1 α ⎪
∂U r ⎪ Gt +1 (1 − α )N ⎪
⎨= 0 if = ϑ ⇔ Z dic = Z demϑ ⎬ (25)
∂τ t ⎪ hit +1 α ⎪
⎪ Gt +1 (1 − α )N ⎪
⎪< 0 if > ϑ ⇔ Z dic > Z demϑ ⎪
⎩ hit +1 α ⎭
and
⎧τ if Z dic < Z demϑ ⎫
⎪⎪ ⎪⎪
τ = ⎨(τ , τ ) if Z dic = Z demϑ ⎬ (26)
⎪ ⎪
⎪⎩τ if Z dic > Z demϑ ⎪⎭

where τ is the lowest possible expropriation rate which is defined by the subsistence

consumption level of the dictator ensuring that ϑτ Nyit ≥ c~ . We assume that the size

17
of the population (N) is relatively large such that the latter condition holds for τ

close to zero. In this case, it follows form (26) and the above threshold level of initial
income ( y ≥ c~ ) that the individual net income ( (1 − τ ) y ) never falls below the
it it

subsistence consumption level c~ . Intuitively, it is never optimal for the dictator to

erode initial individual investments in human capital. In contrast, the dictator
optimally reduces the expropriation rate to a minimum to stimulate private
investments (in human capital) since the latter is a scarce production input.23

If we combine the results of the first two cases, we know that a dictatorship, which is
initially poor ( y ≤ c~ ) grows at a slow and decreasing rate due to positive
it

infrastructure investments until the threshold value for private investments is finally
exceeded ( cit∗ > c~ ). At this stage, the dictatorship is characterized by a relatively high

level of infrastructure capital compared to human capital ( Z dic > Z demϑ ) if the ratio
was relatively balanced initially. It follows that the dictator will drop the expropriation
rate as soon as the private investment threshold is exceeded to promote scarce and
productive private investments instead of relatively unproductive and abundant public
infrastructure investments. As a consequence, the growth rate will temporarily
accelerate in dictatorship as soon as cit∗ > c~ .

The model predictions of high infrastructure investments and unbalanced growth in

poor but stable dictatorial regimes are consistent with the development paths of
several (former) autocratic countries, i.e. China, Vietnam, South Korea, Singapore, etc.
(- compare section 3). Autocratic authorities in these countries invested in large scale
infrastructure projects even though the mayor part of their populations experienced
minimal consumption levels.

Case 3: cit* > c~

∂bit − α (1 − β )
We infer from condition (10) that = yit so that:
∂τ t α (1 − β ) + β

⎧> 0 if Z dic < Zˆ dicϑ

∂U ⎪⎪
r
⎨= 0 if Z dic = Zˆ dicϑ (27)
∂τ t ⎪
⎪⎩< 0 if Z dic
>Z ϑ>Z
ˆ dic ϑ
dem

23
The Chinese economic reforms in the 1980s and 1990s comply with this specific prediction of our model. That
is, the ratio of the Chinese government’s financial income relative to GDP decreased tremendously in 1980s and
1990s: from 31% in 1978 to 10% in 1995 according to Chinese statistics yearbook 2007.

18
 Gt +1 ˆ dic (1 − α )N α (1 − β ) + β
where Z dic ≡ ，Z ≡ > Z dem and
hit +1 α α (1 − β )

⎧τ if Z dic < Zˆ dicϑ ⎫

⎪⎪ ⎪⎪
τ = ⎨(τ , τ ) if Z dic = Zˆ dicϑ ⎬ (28)
⎪ ⎪
⎪⎩τ if Z dic > Zˆ dicϑ ⎪⎭

We are now able to define the long-run expropriation rate ( τ~ ) as the rate that involves

the optimal unique ratio Ẑ dic which maximizes the dictators utility function ( U r )
for a given amount of private investments. This rate is uniquely defined since the
decisions of the dictator over τ (and implicitly btr ) determine a unique

infrastructure capital stock next period and hence the optimal unique ratio Ẑ dic in

the long-run. In particular, if Z dic < Zˆ dicϑ the dictator chooses the highest possible
expropriation rate ( τ ) which entails the highest rate of infrastructure investments. On

the other hand, if Z dic > Zˆ dicϑ she determines the lowest possible expropriation rate

( τ ) entailing zero infrastructure investments. It follows that the ratio of infrastructure

over aggregate human capital adjusts until it amounts to the long-run equilibrium

level Ẑ dic . Thus, the dictator sets the expropriation rate to maximize her utility by
influencing the optimal ratio of public to private capital.

If we combine the results from all three cases, we conclude that the “good
dictatorship” is characterized by a long-run balanced ratio of the infrastructure
relative to the aggregate human capital stock. Hence, GDP-growth follows a unique,
positive long-run balanced growth path as long as the condition for a “good
dictatorship” defined in (23) is satisfied. Moreover, we know from (27) that the
balanced growth path features a higher public to private capital ratio in dictatorship

than in democracy ( Zˆ dic > Z dem ). Thus, the model involves the additional empirical
hypothesis summarized in Proposition III.

Proposition III: Given the preferences and production specifications defined in

(1)-(5), (17)-(19) and a relatively high initial individual income level ( y > c~ ), it
it

follows from (27) that the long-run equilibrium public provision of infrastructure

19
capital is higher in dictatorship than in democracy ( Z dem < Zˆ dic ).24

This result is consistent, for example, with the observed higher infrastructure capital
investments in China relative to India during the last 50 years - compare section 3.

Finally, we combine (1), (10), (22) and (28) and to compute the long-run balanced
growth rate in dictatorship:

g dic =
(1 − α )(1 − β )ψ ⎡1 + τ~ϑAN (Zˆ dic )−α ⎤ − 1 (29)
((1 − α )(1 − β )ψ + β ) ⎢⎣ ⎥⎦

2.3 Comparison of the long-run balanced GDP-growth rates between the two
political regimes

The previous analysis allows us to compare the two long-run equilibrium balanced
growth paths of the two different political regimes for different aggregate income
levels. In contrast to the unique long-run balanced growth rate in dictatorship, the
democratic regime features two possible long-run growth rates. We already know
from Proposition II that the growth rate in dictatorship outweighs the one of the
democratic zero-growth equilibrium if condition (24) is satisfied. In the following, we
compare the two positive long-run balanced growth rates of the two political regimes.
That is, from (16) and (29), we obtain:

ω ⎡β ⎡1 ⎤ α −α β ⎤
g dem > g dic ⇒ τ~ < ⎢ + (1 − α )(1 − β ) + ⎢ − (1 − α )⎥ 1−α ⎥
(30)
ϑ ⎣ψ ⎣ψ ⎦ AN (1 − α ) ⎦
1−α

α
⎛ α (1 − β ) + β ⎞
where ω = ⎜⎜ ⎟⎟ > 1 . This condition is satisfied as long as
⎝ α (1 − β ) ⎠

ω⎛β ⎞
⎜ + (1 − α )(1 − β )⎟⎟ > 1 . First, note that “roving bandits” type of dictatorships
⎜
ϑ ⎝ψ ⎠

(implied by ψ → 0 ) surely satisfy the above condition and thus experience an

inferior long-run growth rate. Second, we are interested in analyzing whether the
“best possible” type of dictatorship, which is defined by ψ = 1 and ϑ = 1 , might

violate the above condition. In this case, the (political) survival of the dictators
offspring is certain and she faces the same enforcement costs as democratic
authorities. Accordingly, the sufficient condition simplifies to ω (1 − α + αβ ) > 1 . The
24
Note that the ratio always remains at the initial level in the democratic zero-growth equilibrium. Hence, a
relatively high initial ratio in democracy would trivially exceed the one in dictatorship.

20
parameter α reflects the weight of human capital in the Cobb-Douglas production
function, while β measures the importance of an individual’s current consumption

relative to the next generation’s income. Assuming a labor share of α = 2 , the

3
growth rate of a democracy outweighs the one of the “best possible” dictatorship if
β > 0.42 . Note that β < 0.42 involves that an individual receives a higher utility
from the income of her offspring than from her own consumption. In other words, she
would invest more than half of her income in the education of her children. This
parameter restriction appears to be quite implausible for the utility function of the
median voter. Thus, the unrestricted positive long-run balanced growth rate of a
relatively rich democratic economy, which features an income level of the median
voter that ensures an interior solution for (6), exceeds the long-run balanced growth
rate of an autocratic economy for reasonable parameter values. 25 The result is
summarized in the following proposition.

Proposition IV: Given the preferences and production specifications defined in

(1)-(5), (17)-(19), it follows from (30) that (i) the long-run growth rate of a relatively
poor dictatorship outperforms the one of a poor democracy if the income of the
median voter is below the subsistence level of consumption ( y ≤ c~ ) and condition it

(24) is satisfied for the dictatorship, while (ii) even the long-run balanced growth rate
of the best performing dictatorship falls below the one of an equally rich democracy if
ω (1 − α + αβ ) > 1 .

Thus, a comparison between the growth performance of dictatorial and democratic

economies depends crucially on the country’s stage of development. The long-run
growth rate in a democratic regime, which is characterized by a relatively high
average income level, exceeds the one of a dictatorship of a comparable income level.
On the other hand, economies that feature low average income levels, according to the
income of the median voter, may experience higher growth in a dictatorial regime. In
particular, relatively poor but large and stable economies, in terms of their population
size and their degree of ethnic diversity, experience higher growth in a dictatorial
regime. This result potentially explains the ambiguous empirical findings associated
with the democracy-growth nexus outlined above. It suggests that the corresponding
relation is non-linear depending on a country’s income level and additional
country-specific characteristics like population size and ethnic diversity.26

25
We note that for the special case of a single input factor of a constant return-to-scale production function
( α = 1 ), condition (30) reduces to τ~ < β . Hence, in this case the growth rate in dictatorship could potentially
1− β
outweigh the one in democracy if the weight of current consumption preferences relative to future is relatively
small.
26
Indeed, Barro (1996) reports some preliminary empirical results which support this non-linear relationship

21
 Figure 1

The dependence of the potential growth paths of the two political regimes are
illustrated in Figure 1. The first income level (Y(1)), defined in (23) and (24),
corresponds to the level that entails a “good dictator” and the second to the income
level that ensures positive investments in education according to (10). Once the latter
is exceeded the democratic economy converges to its unique long-run balanced
growth rate, whereas the specific convergence path in between Y(1) and Y(3) depends
G
on the initial ratio Z 0 = 0 . The increase in the growth rate is higher as soon as
hi 0
Y(2) is exceeded, the more unbalanced this ratio is initially. Proposition IV states that
the growth rate in a dictatorial economy outperform the one in democracy if the
income level of the median voter falls in between Y(1) and Y(2), which are defined in
(23), (24), and (10). The opposite holds if both economies reached their long-run
balanced growth path. Since the income level of the median voter is increasing for
Y>Y(2), there exists a specific income level in between such that economic growth in
a democratic society starts to outperform the one in dictatorship.

3 Democracy vs. Dictatorship: what do the data say?

In the following, we compare the economic performance of different political regimes

based on international (country-) panel data. We identify political regimes using the
Polity IV (2002) codes. Accordingly, we confront the GDP-growth rate as well as the
provision of public capital and social security payments in democracies with their
counterparts in dictatorships of comparable income levels. The aim of this data
analysis is to draw a comparison between the performance of different political
regimes across regions and time periods in order to check for a consistency of our
theoretical approach with systematic variations in the data. We note that, given the
limitations of the available data on developing countries, Table 1, 2 and 3 represent a
rather tentative inspection of potential pivotal differences across the two regimes.

Table 1 contrasts the average growth rates and the pattern of public expenditures in a
country before and after democratization. We restrict attention to countries where the
only experience with regime change over the period under consideration was a major
democratization.27 These are countries that underwent major democratic transitions
according to Polity IV. We also require at least nine years of data before and after
democratization in order to conduct meaningful comparisons of average growth rates.
Twenty-four countries in our sample fit these criteria, and are listed in Table 1.28 We
consider information on the entire duration of a certain political regime. We use the
which depends on a country’s income level.
27
The information on countries that experienced a transition from a democracy to a dictatorship is very limited
since most political regime shifts represent democratization during the last 30 years while the time-series of the
IMF Government Financial Statistics start in 1972.
28
The analysis is based on Rodrik and Wacziarg (2005) (Table 3). The authors analyse the transitional within
country effects of democratization on economic growth.

22
updated (PPP adjusted) income data from Heston et al. (2006) and provide
supplementary information on the structure of a country’s government expenditures
from the IMF Government Financial Statistics in order to shed some light on a
potential source of growth differences. That is, we report the percentage changes in a
country’s public spending to GDP ratio on general public services, transportation and
communication equipment (infrastructure), and social security and welfare after
democratization relative to the previous autocratic era. 29 Unfortunately, these
time-series do not start before 1972. The data exclusively covers low or medium
income countries by the time of democratization as all regime shifts took place in
developing or transition countries.

Table 1 shows that 18 out of 24 countries experienced higher average GDP-growth

rates in the dictatorial regime as compared to the democratic. Moreover, 11 out of 15
countries with a (real) GDP level below 6000 in the year of democratization achieved
a higher growth rate in dictatorship. This descriptive finding supports the theoretical
result summarized in Proposition 2 of our model. The last three columns of Table 1
report variations in public spending on infrastructure and redistribution across the two
political regimes. Our theoretical analysis suggests that democratic regimes invest less
in infrastructure capital, but spend more money on social security and welfare
(redistribution). Indeed, expenditures for general public services relative to GDP fell
below previous ratios after democratization in 9 out of 14 countries with available
data. If we focus on investments in transportation and communication infrastructure
relative to GDP, 9 out of 12 countries exhibited a higher ratio in the dictatorial regime.
In particular, all six countries with available data, which have a level of real GDP
below 6000 by the time of the political shift, reduce their public investments in in-
frastructure substantially (up to 65%) after democratization. 30 Finally, public
spending on social security and welfare display the opposite pattern across political
regimes. That is, in 9 out of 12 cases redistributive spending relative to GDP increased
after democratization. These systematic variations in infrastructure investments and
redistributive policies are consistent with the predictions of Proposition 3 which are
specific to our model: the median voter calls for higher social transfers by the
government at the expense of lower investments in infrastructure capital in democracy
relative to an income-maximizing “good dictator”.

In Table 2, we compare the long-run growth performances of different countries

during the last 50 years. We restrict our attention to relatively less developed countries
and ten year growth intervals. That is, we exclusively confront countries whose
average real GDP during a decade is in the fourth quartile of the overall income
distribution based on the Heston et al. (2006) data. Moreover, we refer to a
democracy/dictatorship if the polity index exceeds 7/-7 and exclude decades with

29
In an extended version of Table 1, we also compare the differences in public spending on education. However,
the results are very similar to the changes in the spending on general public services.
30
We emphasize that transportation or telecommunication services have not been privatized in any of these six
countries. In fact, telecommunication services were e.g. privatized in Spain after democratization, however, overall
public investments in transportation and telecommunication increased.

23
periods of complete collapse of central political authority or occupation by foreign
powers during wartime of at least two years.31 Unfortunately, most poor countries are
ruled by autocratic leaders so that we are left with merely six observations for
democratic regimes. Nevertheless, Table 2 reveals that the average (decade) growth
rates of the least developed dictatorships outperform the ones of the least developed
democracies. The average growth rates amounts to 2.54% in the former subset, but
only 1.09% in the latter. This result is in line with Table 1 and supports the hypothesis
that poor dictatorships often outperform poor democracies in terms of average
long-run GDP-growth.

The comparison of government expenditures or direct measures of infrastructure

capital between least developed countries is very limited due to the absence of
appropriate data. Therefore, we exemplify pivotal differences in the public provision
of infrastructure by confronting corresponding measures from India and China which
represent a stable poor democracy and poor dictatorship, respectively. In addition, the
two regions feature comparable initial income levels.32 The findings are summarized
in Table 3. The average real GDP in India from 1960-2000 amounts to 1338 in
PPP-adjusted $ and exceeds the one of China (1025). Nevertheless, indicators for the
infrastructure capital stock in China outweigh Indian stocks. In particular, the number
of main telephone lines and mobile phones per 1000 workers was on average 21.04 in
China versus 12.40 in India starting from similar initial rates in the 1960s. Moreover,
the power generating capacity amounted to 0.37 Gigawatts per 1000 workers in the
former but only 0.12 in the latter region. In contrast, the average power capacity was
the same in both countries in the 1970s. A qualitative index of the provision of
electrical power, the share of transmission and distribution losses in total electricity
output confirms this result: the loss ratio amounts to merely 0.08 in China, but 0.19 in
India. Yet, the average Indian transportation infrastructure capital stock outperforms
the Chinese: the average length of the total road network is 0.44 km. per sq. km. of
land area in India but only 0.09 km. per sq. km. in China. Apart from differences in
country size most of this difference is due to different initial conditions (British
colonial policy) since 97% of the average percentage deviations in the relative road
stocks between both countries are explained by deviations in initial conditions in the
1960s. Furthermore, the share of paved roads in total roads, a qualitative index of
roads services, was remarkably high in India in the 1960s (40%), but increases only
by 9% since. In contrast, it was historically very low in China (9%), but increases by
14% since.

Summing up, these descriptive statistics suggest that many poor dictatorships
outperform democratic counterparts with respect to average long-run GDP-growth
which is predicted by Proposition II of our model. The IMF Government Financial
31
In particular, we drop the performance of Madagascar in the 1990s (democratic) and of Afghanistan in the
1990s, Congo in the 1990s, Ethiopia in the 1970s, South Korea in the 1980s and Zambia in the 1970s (all
autocratic).
32
Moreover, some country characteristics, e.g. the population size, are comparable among the two economies.
However, there are also important differences, i.e. the degree of ethnic diversification is higher in India than in
China.

24
Statistics data and the comparison of infrastructure indices for China and India
illustrate that differences in (incentives for) infrastructure investments and
redistributive policies under the two political regimes play an important role in this
context. These findings are consistent with our theoretical explanation of growth
differences between democratic and dictatorial regimes, Proposition II-IV, which are
presented in the previous section.

4 Conclusions

In this paper, we compare the long-run growth performances of two economies that
exclusively differ in their political regime in an overlapping-generations framework.
The model features the following key elements. First, individual preferences over
contemporaneous consumption and the income of the next generation allow for the
existence of a corner solution for relatively low individual income levels due to the
existence of a subsistence consumption level. Second, individual productivity depends
on the individual human capital stock and the overall amount of public capital
(infrastructure) in the economy. Both inputs can only be affected dynamically via
investments that pay off in the next generation. In this framework, we demonstrate
that poor democracies exhibit a lower equilibrium balanced growth rate than equally
poor but stable dictatorships since the median voter in the case of democracy
under-invests in productive capital. Intuitively, the low income of the median voter in
a poor democracy involves a relatively low marginal inter-generational utility which
constraints productive investments in future generations. On the other hand, the
higher income level of a dictator, in an otherwise identical economy, implies a higher
marginal inter-generational utility of the decision-maker and hence enhanced
productive investments. If the economy is relatively poor, this effect outweighs the
growth-retarding impact of corruption in dictatorship (dictator’s consumption). Thus,
the dictator of a less developed country may invest in infrastructure while the median
voter calls for redistributive policies to achieve a higher current consumption share.
This redistribution versus public investments policy tradeoff in democratic regimes
has been informally stressed by Barro (1996) and others. In addition, our model
involves a particular threshold value in income such that the growth-reducing impact
of dictatorial consumption (corruption) outweighs the higher (initial) infrastructure
investment. Thereafter, it is shown that the growth rate under democracy strictly
dominates the one in dictatorship. Hence, there exists an intermediate income level
such that economic growth in a democratic society starts to outperform the one in
dictatorship. While the results are at first derived for a homogenous time-invariant
(exogenous) income distribution to keep the dynamic overlapping generations
framework tractable, we also discuss their robustness towards heterogeneous time
variant (endogenous) income distributions. We are able to generate some analogous
results depending on the skewness of an economy’s income distribution.

Moreover, we provide descriptive empirical statistics that suggest that many poor
dictatorships outperform democratic counterparts with respect to average long-run

25
GDP-growth which is predicted by Proposition II of our model. In addition, the IMF
Government Financial Statistics data and the comparison of infrastructure indices for
China and India illustrate that differences in (incentives for) infrastructure
investments and redistributive policies under the two political regimes play an
important role in this context. These findings are consistent with our theoretical
explanation of growth differences between democratic and dictatorial regimes,
Proposition II-IV, presented in the previous section. Summing up, this model shows
that dictatorships, under certain economic conditions, might as well provide good
economic (institutional) outcomes. In particular, the absence of redistribution versus
productive public investment tradeoff might outweigh disadvantages from corruption
in relatively poor economies. In particular, the comparison between the growth
performance of dictatorial and democratic economies depends crucially on the
country’s stage of development, the size of its population, and its degree of ethnic
diversity. This result potentially explains the ambiguous empirical findings associated
with the democracy-growth nexus outlined above. It suggests that the corresponding
relation is non-linear depending on a country’s income level. In fact, Barro (1996)
reports some preliminary empirical results which support this non-linear relationship
with respect to a country’s income level.

Appendix

In the following, we employ the alternative individual preferences of Galor and Moav
(2006) into our model. We show that their alternative preference specification lead to

26
equivalent results in our model. In particular, these preferences involve an explicit solution for the
subsistent income level which entails a poverty trap in democracy. Galor and Moav (2006) assume

the following log-linear utility function: U = β ln(cit ) + (1 − β ) ln( y it + θ ) , where θ denotes

the subsistence consumption level. Hence, the corresponding maximization problem for an
individual in the case of democracy changes as follows:

max U i = β ln(cit ) + (1 − β ) ln(θ + y it +1 ), s.t. (1), (2), (3), (5), (6) ,

τcit , t

where

yi ,t = Ahiα, t Gt1−α
(1)

(2) hit = hit −1 + bit −1 , hi 0 > 0

(3) Gt = Gt −1 + I t −1 , G0 > 0

(4) bit = (1 − τ t ) yit − cit

(5) I t = τ tYt = τ t Nyit .

This results in the following first-order conditions:

∂U i β ⎛ (1 − β ) ⎞⎛ αy it +1 ⎞
: =⎜ ⎟⎜ ⎟ (A1)
∂cit cit ⎜⎝ θ + y it +1 ⎟⎠⎜⎝ hit +1 ⎟⎠

∂U i
: 0=
(1 − β ) ⎡αyit +1 (− y ) + (1 − α ) yit +1 Ny ⎤
⎢ it ⎥ (A2)
∂τ t
it
yit +1 ⎣ hit +1 Gt +1 ⎦

(A1) can be transformed to the following expression:

⎡ β + α (1 − β ) θ ⎤ α (1 − β )(1 − τ t ) yit θ
bit ⎢ + ⎥= − hit − hit (A3)
⎣ β y it +1 ⎦ β yit +1

The above utility function implies that individuals might not invest in future generations if their
income level falls below a certain threshold. The existence of such a threshold can be checked by

setting bit = 0 and τ t = 0 . In particular, setting bit = 0 and τ t = 0 , (A3) must satisfy

∂
≥ 0 which results in the following condition:
∂c c = y

27
 ⎛ θ ⎞
α (1 − β ) y it − hit ⎜⎜1 + ⎟β ≤ 0 (A4)
⎝ y it ⎟⎠

(A4) is a quadratic expression in yit . It can be solved for yit in order to compute the threshold

level of income which involves a poverty trap in democracy (defined by bit = 0 and τ t = 0 ):

βhit ⎛ β hit ⎞ θβhit

2

0 < y it < + ⎜⎜ ⎟⎟ + (A5)

2α (1 − β ) ⎝ 2α (1 − β ) ⎠ α (1 − β )

Hence, the Galor and Moav (2006) preferences lead to an explicit threshold level for income
which implies a poverty trap in democracy. The democratic economy is kept in a poverty trap if
the income level of the median voter satisfies the inequality in (A5). For the initial period, t = 0 ,
the poverty trap condition changes to:

2
α βhi 0 ⎛ βhi 0 ⎞ θβhi 0
Ahi 0 G01−α = yi 0 < + ⎜⎜ ⎟⎟ + (A6)
2α (1 − β ) ⎝ 2α (1 − β ) ⎠ α (1 − β )

Thus, we our utility function in (4) can be derived from the utility function of Galor and Moav
2
β hit ⎛ βhit ⎞ θβhit
(2006) by setting c~ = + ⎜⎜ ⎟⎟ + :
2α (1 − β ) ⎝ 2α (1 − β ) ⎠ α (1 − β )

⎧0 < β < 1 if cit ≥ c~

U = β ln(cit ) + (1 − β ) ln( y it +1 ), ⎨ ,
⎩β = 1 if cit < c~

2
β hit ⎛ βhit ⎞ θβhit
where c~ = + ⎜⎜ ⎟⎟ + .
2α (1 − β ) ⎝ 2α (1 − β ) ⎠ α (1 − β )

Figure 1: Potential growth paths of democracy and dictatorship

28
29
 Table 1: Average growth 10 years before and after (sustained) democratization in
24 countries

Av. Increase Av. Increase Av. Increase

Real public public spending public
GDP in spending transporta spending
PPP public -tion & social security
Year Average Average Average year of services in % communica & welfare in
Country of growth growth growth Dem. of GDP after -tion in % of % of GDP
Dem. before after difference Dem. GDP after Dem. after Dem.
Mali 1991 0.50(29y) 2.52(13y) 2.03 872 - - -
Madagascar 1991 -1.02(30y) -1.59(13y) -0.58 937 -42.9% -15.2% -72.4%
Nepal 1990 2.81(8y) 2.67(10y） -0.14 1106 -59.1% -16.9% 313.6%
Benin 1990 0.60(25y) 1.66(12y) 1.06 1086 - - -
Bangladesh 1991 1.62(15y) 1.52(12y) 0.90 1606 - - -
Honduras 1980 1.08(28y) 0.06(23y) -1.03 2306 - - -
Bolivia 1982 1.18(21y) 0.21(18y) -0.97 2896 - - -
Philippines 1986 1.20(14y) 1.58(16y) 0.38 3016 - - -
Dominican
1978 3.64(11y) 2.75(24y) -0.89 3562 -32.5% -13.9% -8.8%
Rep.
Nicaragua 1990 0.60(27y) -0.87(13y) -1.47 3908 11.8% -12.0% 121.6%
El Salvador 1979 2.01(27y) 0.46(23y) -1.56 4301 - - -
Peru 1978 1.91(9y) -2.03(12y) -3.95 4748 -47.2% - -
Ecuador 1979 6.48(9y) -0.28(24y) -6.76 4901 - - -
Paraguay 1989 2.00(37y) 0.65(13y) -2.64 5175 12.7% -48.3% -22.8%
Panama 1989 2.64(20y) 2.36(14y) -0.27 5876 -64.4% -64.7% 60.8%
Romania 1989 6.21(27y) 0.75(14y) -5.46 6060 275.6% -32.3% 35.9%
Poland 1989 2.80(17y) 2.76(14y) -0.04 6289 - - -
Chile 1989 0.43(15y) 4.33(14y) 3.90 6472 -72.9% - -
Brazil 1985 3.49(24y) 0.59(17y) -2.90 6531 26.1% 10.0% 20.3%
Uruguay 1985 0.97(11y) 2.20(18y) 1.23 6781 -43.0% -0.4% 34.9%
South Korea 1987 6.27(14y) 5.63(16y) -0.64 7375 -21.5% 12.2% 64.4%
Portugal 1974 5.95(23y) 2.65(29y) -3.30 8982 - - -
Hungary 1989 3.36(17y) 1.98(14y) -1.38 10304 -68.2% -40.4％ 35.1%
Spain 1975 6.04(23y) 2.12(25y) -3.92 11614 9.5% 34.5% 49.4%

Sources: IV (2002), Heston et al. (2006), IMF Government Financial Statistics. 带格式的: 法语

We obtain almost identical results if we restrict the number of years to be identical

before and after democratization. The corresponding table is available from the
authors upon request.

30
 Table 2: Democracy vs. Dictatorship: Cross-country growth performances

Dictatorship Democracy
Decade av. growth av. real GDP Decade av. growth av. real GDP
Country Country

Bhutan 70 4.44% 262 Gambia 60 3.16% 786

Bhutan 80 6.15% 481 Gambia 70 -0.20% 921
Bhutan 90 3.43% 673 India 50 1.41% 816
Chad 60 -1.75% 1112 India 60 2.99% 996
China 50 4.82% 382 India 70 1.12% 1203
China 60 0.55% 427 Mongolia 90 -1.94% 1466
China 70 4.63% 587
Congo, Dem. Rep. 70 -1.86% 1343
Congo, Dem. Rep. 80 -2.6% 988
Ethiopia 50 2.02% 377
Ethiopia 60 2.05% 436
Ethiopia 80 -0.67% 470
Ghana 60 12.02% 624
Guinea-Bissau 80 0.44% 623
South Korea. 70 9.58% 644
South Korea 80 6.6 1430
Lesotho 70 4.28% 865
Malawi 60 1.75% 493
Malawi 70 3.12% 613.33
Malawi 80 -0.39% 713.42
Mozambique 70 0.66% 1100
Mozambique 80 0.21% 1005
Nepal 60 0.73% 821
Nepal 70 1.11% 887
Taiwan 50 4.54% 1210
Zambia 70 3.23% 1366
Zambia 80 -0.52% 1265

Average 2.54% 785 1.09% 1031

Sources: IV (2002), Heston et al. (2006).

We refer to a democratic (autocratic) country if the Polity index exceeds 7(-7). We
exclude decades with periods of interregnum of at least 2 years, during which there is
a complete collapse of central political authority or if a country is occupied by foreign
powers during wartime. Excluded countries: democratic: Madagascar 1990s (-1.58%);
autocratic: Afghanistan 1990s (-6.85%), Congo 1990s (-7.01%), Ethiopia 1970s
(0.25%), South Korea 1980s (6.6%), Zambia 1970s (3.23%).

31
 Table 3: Infrastructure capital: China vs. India

Telephone Lines Power Generating 1- Share of transmission Length of total Share of

Average Average
Decade and mobile phones Capacity, Gigawatts and distribution losses road network Paved Roads
growth real GDP
per 1000 workers per 1000 workers in total electricity output km. per sq. km. land in Total Roads

China
60 0.55% 427 3.36 - - 0.05 0.09
70 4.63% 587 3.60 0.08 0.92 0.08 0.13
80 7.61% 1123 5.44 0.14 0.91 0.10 0.19
90 9.49% 2604 71.77 0.27 0.93 0.12 0.23
Average 5.42% 1025 21.04 0.16 0.92 0.09 0.16

India
60 2.99% 996 2.59 0.05 - 0.21 0.40
70 1.12% 1203 5.33 0.08 0.83 0.36 0.40
80 3.65% 1568 9.61 0.14 0.81 0.51 0.47
90 3.53% 2106 32.08 0.22 0.80 0.68 0.49
Average 2.54% 1338 12.40 0.12 0.81 0.44 0.44
Sources: International Telecommunications Union (ITU), United Nations - Energy
Statistics Yearbook and World Bank, International Road Federation. Data on public
spending from IMF Government Financial Statistics is not available for China.

References
Acemoglu, D. (2003), “The form of property rights: Oligarchic vs. democratic
societies”, NBER Working Paper.
Acemoglu, D. and J. Robinson (2000), “Why did the west extend the franchise?
democracy, inequality, and growth in historical perspective”, Quarterly Journal of
Economics (115), 1167–1199.
Acemoglu, D., S. Johnson and J. Robinson (2001), “The colonial origins of
comparative developments: An empirical investigation”, American Economic Review
(91), 1369–1401.
Acemoglu, D., S. Johnson J. Robinson and P. Yared (2005), “Income and
democracy”, NBER Working paper No.11205.
Aghion, P. and P. Bolton (1997), “Wealth inequality as a limit to contracts”, Review
of Economic Studies (64), 151–172.
Barro, R. (1996), “Democracy and growth”, Journal of Economic Growth (1), 1–27.
Barro, R. (2000), “Inequality and growth in a panel of countries”, Journal of
Economic Growth (5), 5–32.

32
Bourguignon F. and Verdier T. (2000) “Oligarchy, democracy, inequality and
growth” Journal of Development Economics (62), 285-313.
Cervellati M., Fortunato P., and Sunde U. (2005) “Hobbes to Rousseau: Inequality,
Institutions and Development”, IZA Discussion paper. 1450
Chinese statistics yearbook 2007
Easterley, W. and R. Levine (1997), “Africa's Growth Tragedy: Policies and Ethnic
Divisions”, Quarterly Journal of Economics, (102), 1203-1250.
Engerman, S. and K. Sokoloff (2003), “Institutional and non-institutional
explanations of economic differences”, NBER Working Paper No. 9989 .
Galor, O. and D. Weil (2000), “Population, technology, and growth: From
malthusian stagnation to the demographic transition and beyond”, American
Economic Review (90), 806–828.
Galor, O. and O. Moav, (2006) “Das Human Kapital: A Theory of the Demise of the
Class Structure” Review of Economics Studies, (73), 85-117
Glaeser, E., R. La Porta, F. Lopez de Silanes and A. Schleifer (2004), “Do
institutions cause growth?”, Journal of Economic Growth (9), 271–303.
Greenwood, J. and B. Jovanovic (1990), “Financial development, growth, and the
distribution of income”, Journal of Political Economy (5), 1076–1107.
Hall, R. and C. Jones (1999), “Why do some countries produce so much more output
than others?”, Quarterly Journal of Economics (114), 83–116.
Heston, A., R. Summers and B. Aten (2006), Penn world table version 6.2.
Polity IV (2002), Political regime characteristics and transitions, 1800-2002.
Jaggers, K. and M.G. Marshall (2000), “Polity iv project”, Center for International
Development and Conflict Management, University of Maryland .
Kaufmann, D., A. Kraay and M. Mastruzzi (2003), “Governance matters iii:
Updated governance indicators for 1996-02”, Working Paper, World Bank
Washington .
Knack, S. and P. Keefer (1995), “Institutions and economic performance:
Cross-country tests using alternative measures”, Economics and Politics (7), 207–227.
McGuire, M.C. and M. Olson (1996), “The economics of autocracy and majority
rule”, Journal of Economic Literature (34), 72–96.
Rodrik, D. and R. Wacziarg (2005), “Do democratic transitions produce bad
economic outcomes?”, AEA Papers and Proceedings.
Shen, L. (2007), “When will a dictator be good?”, Economic Theory 31(2), 343-366
Steger, T. (2000), “Economic growth with subsistence consumption”, Journal of
Development Economics (62), 343–361.
Tavares, J. and R. Wacziarg (2001), “How democracy affects growth”, European
Economic Review (45), 1341–1378.

33