# Do "Clean Hands" Ensure Healthy Growth? Theory and Practice in the Battle Against Corruption

**ABSTRACT** This paper analyzes the existing relationship between economic growth and the monitoring of corruption and examines the possible outcome of the implementation of a State reform in order to weed out corruption. Growth is always higher when monitoring is high and therefore corruption eradicated. But growth declines when monitoring against corruption is not too high, say intermediate, so much that it makes an equilibrium with corruption and little monitoring a more growth-enhancing solution. It is also stressed that when reforms to combat corruption appear to be implausible, they tend to curb most productive investments. The model is estimated using a dynamic panel data approach for Italy. Italy has been plagued by corruption and in the late 80s and early 90s several scandals erupted which led to the well-known "Clean Hands" (Mani pulite) inquiries. Empirical results support the theoretical model.

**0**Bookmarks

**·**

**61**Views

- [Show abstract] [Hide abstract]

**ABSTRACT:**This paper analyzes the relation existing between corruption, monitoring and output in an economy. By solving a dynamic game we prove that equilibrium output is a non-linear upper-hemicontinuous function (MP function) of the monitoring level implemented by the State on corruption, presenting 3 different equilibrium scenarios. According to our model, we analyze the optimal strategy depending on the policy objective of the State and we prove that if the State is budget constrained the optimal policy can lead the economy to an equilibrium with widespread corruption and maximum production.11/2004;

Page 1

Do “Clean Hands” Ensure Healthy Growth?

Theory and Practice in the Battle Against

Corruption∗

Raffaella Coppier

Department of Economic and Financial Institutions

University of Macerata

Mauro Costantini

Department of Economics

University of Vienna

Gustavo Piga

Department of Economics and Institutions

University of Rome Tor Vergata

Abstract

This paper analyzes the existing relationship between economic growth

and the monitoring of corruption and examines the possible outcome

of the implementation of a State reform in order to weed out corrup-

tion. Growth is always higher when monitoring is high and therefore

corruption eradicated. But growth declines when monitoring against

corruption is not too high, say intermediate, so much that it makes

an equilibrium with corruption and little monitoring a more growth-

enhancing solution. It is also stressed that when reforms to combat

corruption appear to be implausible, they tend to curb most produc-

tive investments. The model is estimated using a dynamic panel data

approach for Italy. Italy has been plagued by corruption and in the late

80s and early 90s several scandals erupted which led to the well-known

“Clean Hands” (Mani pulite) inquiries. Empirical results support the

theoretical model.

Keywords: Corruption, Growth, Reform, Panel data.

JEL Classification: C33, D73, K42.

∗The authors would like to thank Francesco Busato, Luca De Benedictis, Gianni De

Fraia, Pietro Reichlin, Maria Cristina Rossi, and seminar participants at the Department

of Economic Sciences at the University of Rome “La Sapienza” and at the Department of

Economic and Financial Institutions at the University of Macerata for their very helpful

suggestions. We would also like to thank Christopher John Norris. A special thanks to

Francesco Nucci. The usual disclaimer applies.

Correspondence address: Gustavo Piga, Department of Economics and Institutions, Uni-

versity of Rome Tor Vergata, Via Columbia 2, 00133, Roma, Italy,

E-mail: gustavo.piga@uniroma2.it

Page 2

1 Introduction

During the last thirty years economists from various fields have contributed

to the analysis of corruption. The first paper that received widespread

attention was published in 1975 (Rose-Ackerman, 1975). Since then many

works have appeared in the economic literature and much attention has been

paid to the relationship between corruption and economic growth. There are

several ways in which corruption may reduce economic growth. Corruption

can act as a tax and thus lower incentives to invest. Corruption could cause

talented people to engage in rent-seeking rather than productive activities.

Corruption may distort the composition of government expenditure as

corrupt politicians could be expected to invest in large non-productive

projects from which considerable bribes can be exacted more easily than

from productive activities.Recent empirical analysis has also provided

evidence of the negative effects of corruption on economic growth (for an

excellent survey see Abed and Gupta, 2002). Only in some cases, it has been

argued that the economic benefits of corruption outweigh its cost. According

to Huntington (1968) “[i]n terms of economic growth, the only thing worse

than a society with a rigid, over-centralized dishonest bureaucracy, is one

with a rigid, over-centralized, honest bureaucracy”.

In this paper, we develop a theoretical endogenous growth model which

incorporates corruption. In our model, the social loss of corruption stems

from the fact that entrepreneurs must bribe a bureaucrat in order to

invest, and consequently devote fewer resources to the accumulation of

capital. Therefore in our model corruption has a negative effect on private

investment. Other models share our framework. Ehrlich and Lui (1999)

claim that individuals have an incentive to compete over the privilege of

becoming bureaucrats (the so-called investment in political capital) since

they obtain economic rents through corruption. This investment in political

capital consumes economic resources which could otherwise be used for

production or investment in human capital. In Del Monte and Papagni

(2001), corruption arises when bureaucrats manage public resources to

produce public goods and services. Corruption reduces the quality of public

infrastructure resulting in a negative effect on economic growth. Barreto

(2000) presents a simple neoclassical endogenous growth model where the

public sector’s monopolistic position is explicitly considered. His findings

indicate that a corruption equilibrium is characterized by lower growth

rates compared to the ideal situation in which public goods are provided

competitively. He also shows that if the public sector is subject to significant

bureaucratic red-tape, all of the agents within the economy may prefer the

corruption equilibrium, as corruption can bypass bureaucratic obstacles.

The novel feature of our paper is a study of the impact of monitoring of

corruption on economic growth. In our theoretical model we derive a non-

linear relationship between the level of monitoring and economic growth,

1

Page 3

as well as between corruption and economic growth. At low monitoring

levels the economy experiences widespread corruption and medium growth

rates; at intermediate monitoring levels no corruption occurs, but low

growth rates are recorded; whereas at high monitoring levels no corruption

occurs and high growth rates are recorded. With reference to this non-

linear relationship, we then consider the effect of a reform aimed at

weeding out corruption. When a reform is announced, economic agents

define an expected probability regarding its permanence in time. If agents

underestimate the probability of monitoring by the State, the effect of such

a reform could be to lead the economy to an unintended equilibrium with

low growth. The non-linear relationship between growth and monitoring

is finally investigated empirically over the period 1980-2003 in Italy. In

order to study this relationship new measures of monitoring are used. Our

empirical evidence supports the conclusions of the model.

The paper is organized as follows: section 2 studies the relationship between

the level of monitoring, corruption, and economic growth. In addition, a

possible outcome of a State reform to curb corruption in terms of growth is

examined. In section 3, empirical implications from the theoretical model

are evaluated. Section 4 concludes.

2Theoretical Model

Let us consider an economy producing a single homogeneous good y. Firms

manufacture y with one input, capital, using one of two technologies with

constant returns to scale: the modern sector technology and the one of the

traditional sector. Each entrepreneur is assumed to have the same quantity

of capital k. The product may be either manufactured for consumption

purposes or for investment purposes.

y = aMk. The entrepreneurs in the modern sector must obtain a licence from

the government to access the technology. In order to obtain such a licence,

an entrepreneur must submit a project to a bureaucrat and this act involves

an implementation cost of sk.1The entrepreneur may access the traditional

sector without any licence being issued. In this case the output is y = aTk.

From this point onwards, it is assumed that (aM−s) > aT> 0, i.e. that the

modern sector is more profitable than the traditional sector. In this economy

there are three types of players: the State, bureaucrats and entrepreneurs.

There is a continuum of bureaucrats and entrepreneurs, and their number

is normalized to 1. Economic agents are risk-neutral. While entrepreneurs

may invest their total capital in the modern sector or in the traditional

one, bureaucrats cannot invest in the production activity, earning a fixed

The modern sector technology is

1The cost of the project submission to the bureaucrat is a function of the investment.

The underlying assumption is that, as the size of the investment grows, the cost for the

entrepreneur’s bureaucratic practices also grows.

2

Page 4

salary w. It is assumed that no arbitrage is possible between the public

and the private sector allowing bureaucrats to become entrepreneurs, even

if their salary w is lower than the entrepreneur’s net return.2Bureaucrats

are corruptible, in the sense that they pursue their own interest, and not

necessarily the one of the State. Bureaucrats are open to bribery as they

issue the licence required to access the modern sector technology to the

entrepreneurs submitting a project. The State controls the bureaucrats in

such a way that they have a probability q (monitoring level) of being detected

if they undertake a corrupt transaction.

In this model, the bureaucrat may decide not to ask for a bribe and to issue

the licence to those entrepreneurs who submit a project, or else to ask for a

bribe (represented by b) in exchange for the licence. Since (aM− s) > aT,

the entrepreneur might find it worthwhile to offer a bribe to the corrupt

bureaucrat with a view to obtaining the necessary licence to access the

modern sector. The bureaucrat is assumed to have both monopolistic power

(i.e. after having submitted the project, the entrepreneur cannot turn to any

other bureaucrat to obtain the licence) and discretional power over granting

the licence (i.e.the bureaucrat may refuse to issue the licence without

being required to provide any explanation). If the bureaucrat is detected

while performing a corrupt transaction, he incurs a cost (either monetary,

moral, or criminal) equal to mk, where m >0; the entrepreneur, if detected,

incurs a cost (either monetary, moral, or criminal) equal to ck, where c >0,

but he is refunded the cost of the bribe paid to the bureaucrat3.

2.1Game Description

In the following, we refer to the entrepreneur payoff by using the superscript

(E) and to the bureaucrat payoff by using the superscript (B). These

represent the first and the second element of the payoff vector ηi,i = 1,2,3,4

respectively. Consider the following three-period game:

Stage 1. At stage one of the game, the entrepreneur decides in which

sector to operate, i.e. whether to invest his capital in the modern or

in the traditional sector. Such a decision is tantamount to the decision

of whether to submit the project to the bureaucrat, considering that

a licence is needed to invest in the modern sector. Project submission

does not automatically result in the bureaucrat issuing a licence, as

he may refuse to grant the licence unless a bribe b is paid. If the

entrepreneur decides not to submit the project (preferring to invest

in the traditional sector instead) the game ends and then the payoff

2This may be assumed since although individuals in the population (bureaucrats) have

a job, they have no access to capital markets, and therefore may not become entrepreneurs.

3See Rose-Ackerman (1999) for details regarding the assumption of a non-constant

punishment function.

3

Page 5

vector is given by:

η1= (aTk,w)(1)

If the entrepreneur decides to submit the project, he asks the

bureaucrat to issue the licence. In this case the game continues to

stage two.

Stage 2. At this stage the bureaucrat, on facing an entrepreneur who has

submitted a project incurring a cost sk, may decide to issue the licence

without asking for a bribe (b = 0).4In this case the game ends and

the payoff vector is given by:

η2= (aMk − sk,w) (2)

Alternatively, if he demands the payment of a bribe (b > 0) from the

entrepreneur before he agrees to issue the licence, the game continues

to stage three.

Stage 3. At stage three the entrepreneur must decide whether to pay a

bribe or not. Should he decide to negotiate the payment of a bribe

with the bureaucrat, the two parties will find the bribe corresponding

to the Nash solution to a bargaining game (bNB). If the entrepreneur

decides not to negotiate with the bureaucrat, the latter will refuse to

issue the licence; thus the game ends with the bureaucrat receiving his

salary and the entrepreneur, after having been denied the licence, will

be left with no other option but to invest in the traditional sector. In

this case the game ends and the corresponding payoff vector is given

by:

η3= (aTk − sk,w)

If the entrepreneur decides to pay the bribe, the expected payoffs

will depend on the probability q with which the bureaucrat and the

entrepreneur are monitored. In this case, the expected payoff vector

is given by:

η4=?(aM− s)k − qck − bNB(1 − q),w − qmk + bNB(1 − q)?

It should be noted that by construction η2is preferred to η3by both agents,

and therefore the bureaucrat will never ask for a bribe which he knows that

the entrepreneur would turn down.

(3)

(4)

4If agents are indifferent about whether to ask for a bribe or not, they will prefer to

be honest.

4

Page 6

2.2 The Solution to the Game

The model described in the previous section well reflects the pervasive

uncertainty which is typically experienced by entrepreneurs when dealing

with the Public Administration. The game may be solved by starting from

the last stage using backward induction, determining the bribe bNB(which

is the Nash solution to a bargaining game). The bribe bNBis the outcome of

a negotiation between the bureaucrat and the entrepreneur (see Appendix

A1 for the proof).

Proposition 2.1. Let q ?= 1.5Then there is a unique non negative bribe

(bNB), as the Nash solution to a bargaining game, given by:

?(aM− aT)k

where µ ≡

α and ω are parameters that can be interpreted as the bargaining strength

measures of the bureaucrat and the entrepreneur respectively.

bNB= µ

(1 − q)

−q(c − m)k

(1 − q)

?

.

(5)

α

α+ωis the share of the surplus that goes to the bureaucrat, and

Let us assume, without loss of generality, that the entrepreneur and the

bureaucrat share the surplus on an equal basis. Thus we have a standard

Nash case, when α = ω = 1 and the entrepreneur and the bureaucrat receive

equal shares. Hence the bribe is equal to:

?(aM− aT)k

bNB=1

2 (1 − q)

−q(c − m)k

(1 − q)

?

.

(6)

2.2.1Static equilibrium

The game is solved by means of backward induction starting from the

last stage and the solution is formalized by the following proposition (see

Appendix A2).We here focus on the case where parameters allow the

greatest number of equilibria depending on the level of monitoring by the

State6.

Proposition 2.2. Define q2 =

q1> q2. Then, if q2≥ 0, q1≤ 1 and (c + m) > 2s:

(a) If q ∈ [0,q2] then the equilibrium payoff vector is:

?(aM+ aT)k

(aM−aT)

(c+m)

−

2s

(c+m)and q1 =

(aM−aT)

(c+m)

with

η4=

2

− sk −q(c + m)k

2

,w +(aM− aT)k

2

−q(c + m)k

2

?

(7)

this is the payoff vector connected to equilibrium C (see below);

5If q = 1 this stage of the game is never reached.

6In the Appendix A3 we show the results under all parameter conditions.

5

Page 7

(b) if q2< q < q1the equilibrium payoff vector is:

η1= (aTk,w)(8)

this is the payoff vector connected to equilibrium B (see below);

(c) If q ∈ [q1,1] the equilibrium payoff vector is:

η2= ((aM− s)k,w)(9)

this is the payoff vector connected to equilibrium A (see below).

The previous proposition shows that, depending on the parameter values,

one of the three perfect Nash equilibria is obtained in the subgames:

• Equilibrium C: corruption and high output. When 0 ≤ q ≤ q2, i.e.

if the monitoring level is low enough, the entrepreneur will enter the

modern sector and will be asked to pay a bribe by the bureaucrat.

Monitoring intensity is so low that the difference in gross profits,

(aM− aT)k, between the modern and the traditional sector is high

enough to outweigh a (relatively low) expected cost of corruption and

the cost of the project.

• Equilibrium B: no corruption and low output. When q2< q < q1, i.e.

if the monitoring level is intermediate, the entrepreneur will not enter

the modern sector and therefore will not ask for a licence. Monitoring

intensity is not low enough for the entrepreneur to justify paying for the

cost of the project along with the additional expected cost of paying a

bribe. The difference in gross profits between modern and traditional

sector does not compensate for the expected cost of corruption plus

the cost of the project. Furthermore, monitoring intensity is not of a

high enough level to deter the bureaucrat from asking a bribe in case

the entrepreneur were to pay the cost of the project.

• Equilibrium A: no corruption and high output. When q ≥ q1, i.e. if

the monitoring level is high enough, the level of monitoring intensity by

the State is so high that the entrepreneur would turn down a request

for a bribe even after having paid the (sunk) cost of submitting a

project. Realising this fact, the bureaucrat will refrain from asking

for a bribe to issue the licence. Thus the entrepreneur will enter the

modern sector and will not be asked for a bribe by the bureaucrat.

Notice that in equilibrium B there is no corruption but low output compared

to equilibrium C where corruption is at its highest, but output is higher.

Should a State wish to lead the economy towards one of these three viable

equilibria by employing a certain level of monitoring, it would realise that

equilibria A and C imply a greater output than equilibrium B. Equilibria

6

Page 8

A and C allow the same output to be obtained, even though they are

considerably different from one another in terms of level of corruption (which

is greatest in C and nonexistent in A).

From a static perspective, the equilibrium A is better than equilibrium C

which implies the same output of equilibrium A but is characterized by

widespread corruption, entailing a higher cost, summarized by parameters c

and m. A is also better than B, while B and C cannot be ranked a priori.

2.2.2Dynamic equilibrium

Following the work of Del Monte and Papagni (2007), we expand the game

perspective in order to examine the dynamic consequences of corruption

on economic growth and hence on investment. Satisfaction is derived from

consumption according to a simple constant elasticity utility function:

U =C1−σ− 1

1 − σ

Each entrepreneur maximizes utility over an infinite period of time subject

to a budget constraint. This problem is formalized as:

?∞

sub

•

k = ηE

where C is consumption, ρ is the discount rate over time, ηE

entrepreneur’s payoff.

Since ηE

each of the three cases. In the equilibrium with corruption (equilibrium C),

the entrepreneur’s payoff is:

?(aM+ aT)k

thus the constraint is:

?(aM+ aT)k

The Hamiltonian function is:

?(aM+ aT)k

where λ is a costate variable. Optimization provides the following first-order

conditions:

e−ρtC−σ− λ = 0

max

c∈?+

0

e−ρtU(C)dt

i− C

i

is the

iis different in each of the three equilibria, the problem is solved for

ηE

4=

2

− sk −q(c + m)k

2

?

•

k =

2

− sk −q(c + m)k

2

?

− C

H = e−ρtC1−σ− 1

1 − σ

+ λ

2

− sk −q(c + m)k

2

− C

?

(10)

7

Page 9

and

•

λ= −λ

?(aM+ aT)

2

− s −q(c + m)

2

?

(11)

By differentiating the first condition (10) with respect to time and

substituting it into the second condition (11), the consumption growth rate

is obtained:

γC

σ

2

In equilibrium A, the entrepreneur’s payoff is:

C=1

?(aM+ aT)

− s −q(c + m)

2

− ρ

?

ηE

2= aMk − sk

In this case, optimization provides the first-order conditions that allow the

corresponding consumption growth rate to be obtained:

γC

A=1

σ[aM− s − ρ]

In equilibrium B, the entrepreneur’s payoff is:

ηE

1= aTk

and the corresponding consumption growth rate is:

γC

B=1

σ[aT− ρ]

It should be noted that:

γC

A> γC

C> γC

B

i.e.

highest consumption growth rate; in equilibrium C (pervasive corruption,

low monitoring) the consumption growth rate is intermediate; and finally

in equilibrium B (no corruption, intermediate monitoring level) the

entrepreneur invests in the traditional sector, with low profits, low

accumulation of capital and a low growth rate. Furthermore, it can be shown

that capital and income also have the same growth rate as consumption.

Therefore, of the three equilibria, from a dynamic viewpoint, equilibrium

A is the most conducive to economic growth. This is shown in Figure 1

in terms of monitoring level and growth rate: equilibrium A (high-level

monitoring without corruption) produces the highest growth rate since the

entrepreneurs, who are investing in the modern sector without paying bribes,

are able to generate greater accumulation of capital; in equilibrium C the

growth rate is intermediate since although the entrepreneur manages to

invest in the modern sector, he must pay bribes in order to do so and ends

up accumulating less; finally in equilibrium B the entrepreneur invests in the

the equilibrium A (no corruption, high-level monitoring) has the

8

Page 10

Figure 1: Monitoring and Growth rate.

q2

?

q1

?

C

?

B

Equilibrium C Equilibrium B Equilibrium A

?

A

9

Page 11

traditional sector, with low revenues and low accumulation of capital. Thus

a non-linear U-shaped relationship between the monitoring of corruption

and economic growth is obtained which shall be tested empirically in the

next section.

Before doing that, notice that a farsighted State, with long-term

objectives, will be encouraged to implement a reform to weed out corruption

by leading the economy from equilibrium C to equilibrium A, by means of

increasing the level of monitoring q. While this is true also for the static

context of the previous section, it should be noted the added advantage that

arises in a dynamic context; not only do the costs of corruption decline, but

also investment, consumption and output are higher when the State raises

the monitoring intensity to a sufficiently high level. Let us thus assume that:

a) The economy is at an equilibrium characterized by widespread

corruption, i.e. at equilibrium C, where q∗is the current monitoring level

with 0 ≤ q∗≤ q2;

b) The State announces that a higher monitoring level will be

implemented, and one by which the probability of being monitored while

performing a corrupt transaction increases from q∗to q1, with 0 ≤ q∗≤ q2.

Hence we assume that the economy initially has a high level of corruption

and economic growth is at an intermediate level, and that the State

announces a reform designed to increase growth and eradicate corruption (to

achieve equilibrium A). Since q1is the minimum monitoring level required

for the economy to achieve equilibrium A, the State will raise the probability

of being monitored to q1.

We assume that agents will formulate expectations regarding whether or not

the reform will last over time. Economic agents are assumed to estimate that

there is a given probability (1−π) that monitoring will revert to the previous

level q∗. It is further assumed that the State is unaware of this belief held by

economic agents. When a reform is announced, economic agents evaluate the

probability of being detected (qe), weighted according to their assessment of

whether or not the reform will last over time. The probability expected by

the operators is qewith qe< q1and equal to qe= q1π + q∗(1 − π). In this

case, even if the State intends to increase monitoring to steer the economy

to equilibrium A, operators will expect a lower value of monitoring, qe. If

this value is such that:

(aM− aT) − 2s

(c + m)

< qe<(aM− aT)

(c + m)

the economy ends up at equilibrium B. In terms of economic growth this

outcome is not only worse than equilibrium A to which the State aspires to,

but is also worse than the baseline equilibrium C. In fact, equilibrium B is

characterized by a growth rate lower not only of equilibrium A, but also of

equilibrium C.

Although the reform was intended to foster growth by eradicating

10

Page 12

corruption, due to its lack of credibility it induces entrepreneurs to invest in

the traditional sector rather than in the more innovative modern sector.

Thus the level of economic growth was higher before the reform, when

corruption was prevalent. This feature of the model reflects the oft-heard

argument that “corruption greases the wheels of growth”, but also qualifies

it. Only if the State lacks complete control of the monitoring technology can

such prescription apply. A credible, appropriate reform will always foster

growth by curbing corruption.

3 Empirical Analysis

The relatively recent Italian nationwide Mani Pulite (“Clean Hands”)

scandal, in conjunction with judicial authorities implementing greater

levels of monitoring as a consequence, drastically affected Italy’s economic

environment. This Italian experience lends itself naturally to verify the

impact of the announcement and implementation of a “monitoring reform”

on corruption and growth.7This section aims to empirically investigate the

relationship between the level of monitoring of corruption and economic

growth in Italy.The non-linear character of the relationship between

monitoring level and the growth rate of income is formalized by using

an empirical specification reflecting a parabolic relation between these

two variables.The theoretical model is tested using new measures of

monitoring.8

3.1Data

The empirical analysis is based on annual data from Italian regions over

the period 1980-2003.With the exception of monitoring and human

capital variables, the annual data are drawn from the Prometeia Regional

Accounting data-set (courtesy of ISAE). The data relating to monitoring

are selected from ISTAT and the Ministero dell’Economia e delle Finanze,

and data regarding human capital are drawn from the Costantini-Destefanis

(2009) data-set. Appendix B provides a detailed description of the variables

and their sources. The descriptive statistics of the variables are found in

7Mani pulite (Italian for “Clean Hands”) was a nationwide Italian judicial investigation

into political corruption, held in the 1990s. As Della Porta and Vannucci (1999) said “In

Italy the history of corruption does not begin (let alone end) on February 17, 1992 (the

official date that Clean Hands began). What starts at that point are the extraordinary

events of public exposure of corruption, a scandal affecting the highest levels of the political

and economic system, causing the most serious political crisis of the Italian Republic”.

The corruption system that is uncovered by these investigations is usually referred to as

Tangentopoli, or “bribesville”.

8In recent years several authors have investigated the causes and consequences of

corruption in Italy (Del Monte and Papagni, 2001; Golden and Picci, 2005; Del Monte

and Papagni, 2007)

11

Page 13

Appendix C.

As regards the monitoring variable, three different measures are provided

with a view to studying the effects of monitoring on economic growth. The

first is based on the number of corruption crimes and is denoted as M1.

The second and the third measures use the number of pertinent judges

and policemen as proxies to indicate how much of the State resources are

allocated to fighting corruption related crimes. These two measures are

denoted as M2 and M3. The first measure (M1) is the ratio between the

number of corruption crimes detected and the estimated total number of

corruption crimes (see Appendix D). This is an ex-post variable since it only

considers the results of State control activity. Therefore this index expresses

the effectiveness of the monitoring activity, i.e. the monitoring which leads

to the corrupt bureaucrat being successfully charged. The second and third

measures are based respectively on the number of judges assigned to penal

law cases (M2) and on the number of policemen employed in the investigation

of corruption related crimes (M3). Incentives for corruption increase as the

probability of being caught and punished decreases and this probability is

positively dependent on the actions of judges and policemen. These two

proxies are of an ex-ante nature, since they allow us to assess the level of

monitoring implemented by the State.

3.2Estimation Methods

The specification of the basic estimated equation corresponds to a reduced

form so as to evaluate the implications of the theoretical model. Following

the work of Levine and Renelt (1992), a degree of convergence on the most

appropriate empirical specification for modeling growth has occurred (see

Temple, 1999). Our base specification is fundamentally a “Levine-Renelt”

one with the addition of the monitoring variable. We differ in that since

our estimating model is dynamic rather than static (see Greenaway et al.,

1998, 2002) and the growth rate is included as a regressor (lagged by one

period). This specification has an obvious intuitive appeal in that it models

growth in a dynamic context. However, when a lagged dependent variable

is included in the regression, a correlation between the error term and this

variables may be found. To provide consistent estimates, an instrumental

variable procedure is adopted (see discussion on the GMM system estimator

below). The regression equation is:

gyit= β1gyit−1+ β2lnyit−1+ β3ln(monitorit−1) + β4(ln(monitorit−1))2+

+ β5invit+ β6conpait+ β7hit+ eit,

(12)

where gyitis the growth rate of the per capita income at 2000 constant

prices, gyit−1is the lagged growth rate of the per-capita income, lnyit−1is

the logarithm of lagged value of the per capita income level, ln(monitorit−1)

12

Page 14

is the log monitoring level delayed by one period, (ln(monitorit−1))2is the

square of the logarithm of the monitoring variable lagged by one period,

invitis the share of investment in GDP, conpaitis public consumption over

GDP and hitis the stock of human capital.9The term eitrepresents the

error term, which is assumed to be IID. The index i refers to the cross-

section dimension (regions) and the index t to the time dimension. The

share of investment over GDP and the level of public consumption over GDP

are important control variables.10The “monitoring” variable is included in

the equation with a delay of one period for two reasons: firstly, because

changes in the monitoring level are very likely to require some time before

they influence the agents’ decisions;11and secondly, any distortions due to

simultaneity, resulting from the possible endogeneity of the “monitoring”

variable, need to be mitigated, since a higher growth rate may result in

more tax revenue and therefore more resources being allocated to monitoring

activity.

Equation (12) is estimated using a GMM system estimator proposed by

Arellano and Bond (1991) and Blundell and Bond (1998). The first step of

the estimation procedure is to ensure that the model is either specified with

fixed effects or with random effects. In our case, the fixed effect model has

been preferred as the Hausman test results suggest (35.32, p-value 0.00).

In the dynamic panel, the OLS estimator is biased and inconsistent, even

though the error term is not serially correlated (see Hsiao, 2003). The LSDV

(least square dummy variable) model with a lagged dependent variable

generates biased estimates when the time dimension of the panel is small

(see Nickell, 1981), while when the time dimension is large (up to 30

observations), a corrected LSDV estimator seems to work better with respect

to the OLS, IV and GMM estimators, with the GMM estimators as a second

best solution (see Judson and Owen, 1999). However, the correction for

LSDV derived by Bun and Kiviet (2003) for balanced panels only works

with exogenous explanatory variables. For this reason equation (12) was

estimated with a system of GMM estimators developed by Arellano and

Bond (1991) and Blundell and Bond (1998). A finite sample correction for

the two-step covariance matrix derived by Windmeijer (2005) is used in the

estimation of the equation (12). This correction makes the robust two-step

procedure more efficient than the one-step procedure, especially for system

GMM. The estimation results are summarized in the next section.

9Details of data-set construction are found in Costantini and Destefanis (2009).

10See Barro (1991) and Levine and Renelt (1992).

11Del Monte and Papagni (2001) use two lags for the corruption variable in the economic

growth regression equation for Italian regions.

13

Page 15

3.3Results

The estimation results are reported in Table 1.

instruments used are “yit”, “gyit”, monitor, monitor2, Inv and conpa lagged

for two periods. Furthermore two period dummies are used as instruments

in all regressions as they are related to the monitoring variable. Dummy1

relates to the 90s decade and includes the entry into force of law n. 86 (26th

of April, 1990) which substantially modified the provisions of the penal code

with regard to corruption. Dummy2 takes account of the beginning of the

“Clean hands” investigation (1992).

The estimated regression coefficients of the square of the logarithm of

the monitoring variables are all positive (see Table 1). This confirms the

existence of a parabola-shaped relationship with the concavity upwards as

predicted by the theoretical model.

statistically significant. As regards interpreting the effect of the square of

the logarithm of the monitoring variable on economic growth, since the

coefficient of the logarithm of the monitoring is negative and the coefficient

of the square of the logarithm of monitoring is positive, this equation implies

that at low values of monitoring additional monitoring units have a negative

effect on the dependent variable. At some point, the effect becomes positive,

and the quadratic shape indicates that the impact of monitoring on growth

is increasing as the monitoring itself increases. The sign of the parameter β1

is positive in all cases, signifying a positive correlation between the lagged

delayed growth rate and the income growth rate at time t. The associated

t-statistics are also significant.

The coefficient on the lnyit is used to test the convergence hypothesis.12

A negative sign denotes conditional convergence of growth rates. In our

estimation, the sign of the parameter β2is negative in all cases and varies

from -0.123 to -0.165.These estimated coefficients are also statistically

significant. As regards the investment/GDP ratio (Inv) variable, the

estimated coefficients are all positive and statistically significant. The values

of these coefficients vary from 0.143 to 0.165. These results would seem to

be in line with the literature concerning growth models (see e.g. Levine

and Renelt, 1992) and similar findings are also found in other studies of

Italian regions (see Auteri and Costantini, 2004).

public consumption variable, positive coefficients are found in all cases. The

estimated coefficient values vary from 0.188 to 0.247 and are all statistically

significant. Del Monte and Papagni (2007) found similar results, although

their evidence of a positive impact of public consumption on economic

growth is weaker. As regards the human capital variable, a positive and

statistically significant effect on economic growth is also found. The level

of education has a crucial impact on growth as it determines the economy’s

For all equations, the

The estimated coefficients are also

With respect to the

12Jones and Manuelli (1990) and Kelly (1992) are early examples of endogenous growth

models compatible with β − convergence.

14

#### View other sources

#### Hide other sources

- Available from Gustavo Piga · May 17, 2014
- Available from univpm.it