1

February 22nd 2020

The Transition of Corruption Institutions and dynamics

Martin Paldam, Aarhus University, Denmark1

Abstract: The data for corruption/honesty and income have a correlation of about 0.75, and a

typical transition path. However, the correlation of the growth rate and honesty is negative.

Thus, the short- and long-run findings are in contradiction. It is shown that the contradiction

lasts a dozen years. The transition of corruption happens relatively late in the process of

development and works through earlier changes in institutions. Countries with high and stable

wealth and modern institutions get low corruption. Variables for the economic and political

system explain as much as income, but system changes and uncertainty increase corruption.

This is a second contradiction much like the first one. The two system variables have transitions

too. To identify the non-spurious part of the relations, the D-index is defined as the difference

between the corruption index and the transition path; it reduces the relations to the system

variables, but they do not disappear.

Keywords: Corruption; transition; institutions

Jel code: D73, K42, P48

Acknowledgement: The paper belongs to a project that is joint work with Erich Gundlach, and also this paper owes much to Erich. I have tried to make the paper independently readable, but I shall not repeat what is already published. The paper has been presented at the Ariel (2019) and (soon) the Münster (2020) conferences of political economy, and (soon) at the (US) Public Choice (2020) conference. I want to thank the discussants at these conferences for fine comments, and the two referees for constructive advice.

1. Department of Economics and Business, Fuglesangs Allé 4, DK-8210 Aarhus V, Denmark. E-mail: mpaldam@econ.au.dk. URL: http://www.martin.paldam.dk. Corresponding author.

2

1. Introduction: two contradictions

This paper uses the corruption index (that increases with honesty) to demonstrate four points:

(p1) The index has a large positive correlation to income, giving a beautiful transition curve,

where the change from corruption to honesty is explained by income.

(p2) The index has a small negative correlation to economic growth, where income variation

explains rising corruption. Seen together, (p1) and (p2) are the first contradiction.

(p3) The Transition of Corruption happens rather late in the development. It is being pushed

by the prior transitions in both the economic and political systems.

(p4) Corruption also rises due to institutional changes/uncertainty and crises that occur

during the transitions of the institutions. (p3) and (p4) are the second contradiction.

The two contradictions both say that a good society – with high and stable wealth and consoli-

dated modern institutions – gets low corruption, but to get there requires many changes that

increase short-run corruption.

The paper argues that the correlation (p1) is caused by the Transition of Corruption

when countries move from low to high income.2 This causal interpretation is controversial, as

many have the reverse belief, i.e. that corruption is a barrier to development. It is known as the

sand theory, where corruption is seen as sand in the machine of development. It argues against

the sand theory that the Transition of Corruption happens late in the development.

I.1 A preview and definitions of variables and concepts

Table 1 is a condensed preview of the results, as regards the first two points (p1 and p2), in the

form of four correlation matrices. The left-hand panel gives results for country averages, and

the right-hand panel gives results for unified annual data. Results above the diagonals are for

all data, while results below the diagonal are for the Main sample of non-OPEC countries. The

four (p1) cells all report large correlations of corruption, T, and income, y. Cells (p2) report four

negative correlations of T and growth, g.3 (p1) must be generated by some relation(s) in the

short run, but (p2) surely fails to aggregate to (p1).

2. The total amount of money changing hands due to corruption is small relative to GDP in all countries. The causal direction found is thus in accordance with the tail wagging model: A dog can wag its tail, but not the other way around, though there might be repercussions where a vigorous wagging moves the dog a little. 3. This is consistent with most prior research; see the meta-study by Ugur (2016). However, the new study by Gründler and Potrafke (2019) finds a small positive coefficient in a model with many lags.

3

Table 1. Some correlations of country averages and all observations, 1995-2016

Country averages N = 166 and 151 All observations N = 2,977 and 2,730 T D y g T D y g T, corruption - 0.60 0.73 (p1) -0.16 (p2) - 0.64 0.73 (p1) -0.08 (p2) D, new index 0.65 - -0.04 -0.08 0.67 - 0.02 -0.00 y, income 0.78 (p1) 0.12 - -0.04 0.77 (p1) 0.12 - -0.07 g, growth -0.17 (p2) -0.17 0.00 - -0.09 (p2) -0.01 -0.06 -

Note: Results over the diagonal are for all countries, while results below the diagonal are for the Main sample. Significant correlations are bolded. The D-index measures corruption relative to the transition path.

A central concept in the analysis is the transition. For any variable, X, the transition is

the average path ΠX(y) as a function of income, y, when countries change from the traditional

steady state to the modern one. Normally the levels of X are stable at the two steady states and

quite different, hence the transition. It can be studied in long time-series or wide cross-country

samples – for corruption only the latter is available. Π-curves are estimated on unified data by

kernel regressions, K(y, bw), where bw is the bandwidth. This paper looks at the transition of

corruption, and relies on prior findings for the transitions of the political and economic systems.

Table 2. Variables used in the paper – for easy reference Variable Definition and source. The indices, it, are for country and time. Unified data has index, j a)

National accounts development indicators, from the Maddison project GDP Gross National Product, in fixed PPP prices gdp GDP per capita. The cgdppc series are used as they are made to analyze time series yit Income, the natural logarithm to gdp git Growth of gdp, the five-year average is termed g5 Variables from Transparency International’s Corruption Index

Tit The T-index [0, 10] for corruption to honesty. It rises when corruption falls. Dit The deviation of the T-index from the transition path: D = T − ΠT ΠT(yj) Transition of Corruption. Estimated by the kernel KT(y, bw), where ∂ΠT/∂y > 0

Variables from the Polity index; see Paldam and Gundlach (2018) and Paldam (2020) b) Pit The P-index [−10, +19], for authoritarian regimes to democratic ones Vit

P The average numerical change in P per year c) Zit

P The fraction of years with P = 0, i.e. anarchy or temporary foreign domination ΠP(yj) Democratic Transition. Estimated by the kernel KP(y, bw), where ∂ΠP/∂y > 0

Variables from the Fraser index of economic freedom; see Paldam (2020) b) Fit The index [0, 10], for no economic freedom to the laissez faire Vit

F The average numerical change in F per year c) ΠF(yj) Transition of the Economic System. Estimated by the kernel KF(y, bw), where ∂ΠF/∂y > 0

Notes: (a) Unified data merge the two dimensions t and i. The index j gives the order of the data. Transitions, Π(y), are estimated on unified data by kernel regressions, K(y, bw), where bw is the bandwidth. (b) The transitions in P and F is analyzed and updated in Paldam (2020) based on prior papers. (c) The formula is / ( 1)eX

itt sV X e s

== ∆ − +∑ , where X is either P or F from s (start) to e (end).

4

This means that transitions are the consequence of the whole confluent complex of

development, including the advances in education, health etc. Income is an aggregate for the

whole complex, and it is the best proxy for development, except in special cases such as the

OPEC-countries that have had much larger increases in income than in development.

Section 2 discusses three theories connecting corruption to income. (i) The demand

theory: Honesty is a ‘good’ with a positive income elasticity. (ii) The factor theory: Honesty is

a factor of production. The third is a long-run theory. (iii) The transition theory claims that

development causes honesty.

Section 3 estimates the long-run transition, i.e. T = ΠT(y) ≈ KT(y, bw), and demonstrates

that the curve is robust. In the shorter run, the relation between y and T has complex dynamics,

giving the contradiction that lasts a dozen years. The paper argues that causality is from income

to the T-index in accordance with the demand and the transition theories. A new D-index

measures corruption relative to the transition path, i.e. are countries too corrupt or too honest

for their income level? Table 1 shows that when the D-variable is used, the positive correlation

to income vanishes as it should, and the negative growth effect decreases, so the contradiction

gets smaller, but it does not disappear. The contradiction suggests that the positive relation of

income and corruption is indirect, so that it works through some other variables. The paper

discusses if these variables are institutional using two main institutional indices, P for the

political system and F for the economic system, as defined in Table 2.

Section 4 analyzes the relation of T and D to the two system-indices and finds (p3) that

they have a strong effect on the T-index. However, both indices have transitions, as analyzed

elsewhere (see Table 2), so their relation to the T-index is partly spurious. The D-index should

take out the spuriousness. This reduce the size of the coefficients, but they are still significant.

It is also important that the transitions in the two system variables happen earlier than the

transition of corruption. The connections (p4) between T and D and changes in the two system

indices is found to be negative. It is argued that variations and uncertainty of institutions

increase corruption.

Section 5 gives six examples showing the short-run problems. The first three are to

country pairs that are similar in most respects, but where one has much more institutional

instability and more corruption. The last three are to countries with unusually large crises, where

corruption rises. Section 6 concludes.

A net-appendix is posted at http://martin.paldam.dk/Papers/GT-Main/T-appendix.pdf.

References to tables and figures in the appendix have an A before the number, such as Table

A1.

5

2. Notes on the literature

Scattered corruption data started to appear in the 1980s, and they were compiled to an aggregate

index in the mid-1990s by the NGO Transparency International. The old literature before that

was theory based on anecdotal evidence.4 After data became available, a new literature

emerged.5 Two large books survey these literatures and republish the main papers. They are:

Heidenheimer et al. (1989) for the old literature, and Dutta and Aidt (2016) for the new.

The first question in the old literature is how to delimit corruption from other types of

fraud and rent seeking. A number of definitions have been proposed. The paper uses Transpa-

rency’s definition: Corruption is the abuse of entrusted power for private gain. This definition

implies a principal agent framework, with an agent who deals with a third party. Corruption

occurs when the agent colludes with the third party to defraud the principal. The longest chains

of agents and sub-agents exist in the public sector. Hence, it is particularly prone to corruption.

The first question in the new literature is the quality of the measurement: The T-index

aggregates a number of primary indices of perceptions of corruption by a calibration method

developed by Johann Graf Lambsdorff; see the appendix of his book from 2007.6 Aidt (2003)

gives a survey of the early measurement discussion; see also Gutman and Paldam (2020). The

compilation method for the T-index has changed over time, notably in 2012; see Gründler and

Potrafke (2019). However, when the annual cross-country correlations cor(y, T) and cor(g, T)

are calculated, they have no kink in 2012; see Figure A1 (appendix). Thus, the operational

solution is to note that the index has a measurement error of almost one T-point. The T-index

contains a complex process of residual correlation that is partly caused by measurement, which

has changed over time. Until section 5, conclusions are based on cross-country averages over

N countries, where N > 100, so the error is divided by 10.N >

2.1 Three theories about the relation of corruption, T, income, y, and growth, g

The large correlation (p1) of income, y, and corruption, T, can be explained both from the

consumption and the production side by two simple mono-causal theories.

4. The old literature suffered from politeness. Though it was widely known, it was considered impolite to mention that the level of corruption is higher in poor countries. Consequently, most papers in Heidenheimer dealt with the USA, and the 1,776 pages of volume 1 of the Handbook of Development (Chenery and Srinivasan 1988-89) did not mention corruption. This changed after data appeared, as seen already from the title of Dutta and Aidt (2016). 5. Treisman (2000) and Paldam (2001, 2002) started the cross-country approach independently in 1999, where the T data covered 100 countries, with N = 336. In 2019, the data cover 188 countries and N = 3,156. 6. The calibration means that the 38 countries covered all years have virtually the same average score throughout, but the average for all countries falls due to the gradual inclusion of more low- and middle-income countries.

6

The demand theory sees honesty as an intangible good with a positive income elasticity.

It speaks for this theory that it can be extended to a whole family of intangibles, such as

democracy, generalized trust and various cultural goods. They are ‘nice’ to have, but not really

necessary, so the consumption of these goods increases when income rises. Most poor countries

are weak on honesty and democracy, and have few art museums. The consumption of other

intangibles, such as religion, decreases when income rises. Thus, intangibles may have both

positive and negative income elasticities. Such goods are imprecisely measured, and not sold

on the market, and their prices are difficult to impute, though many attempts have been made.

Thus, the parallel to goods is a bit of a construct. In this theory the causality is: .y T⇒

The factor theory sees honesty as a factor of production that saves time and efforts in

all transactions; it sees bribes as a cost, and notes the costs of the tricks agents use to extract

bribes. This will be further discussed in the next section. Here, causality is the reverse, and has

two links: .T g y⇒ ⇒

The transition theory sees the change from corruption to honesty is a consequence of

development. This is a reduced form relation with the causal direction .y T⇒ At the end of the

section, it is discussed if it can be modeled by an indirect mechanism that includes point (p2).

2.2 The causality controversy and the sand vs grease parables

The demand and the transition theories see corruption as a social ill that vanishes over time as

countries develop, while the factor theory provides a double argument to fight corruption. It is

not only a social ill in itself, but also sand in the machine of development. Many authors,

notably Lambsdorff (2007), stress this theory and provide some evidence. The key argument is

that corruption is an extra cost of transactions (and thus production). This should give a positive

correlation of T and growth, contrary to the evidence (p2) that cor(T, g) < 0.

Given (p2), it is possible that corruption works as grease in the machine, increasing

efficiency. Thus, in a growth perspective corruption is either sand or grease in the machine.7 In

other words, a briber may see the bribe as a cost or as a cost-saving device. Especially as regards

public regulations, it is easy to come up with examples supporting either view. The examples

hinge upon externalities:

Many regulations improve welfare, so corruption reduces the improvement. Examples

are compulsory inoculation programs to eradicate epidemic diseases, or regulations reducing

7. The sand theory is, as mentioned, much more popular. The grease theory goes back to Leff (1964).

7

air pollution, etc. Even if corruption allows the individual a welfare improvement by circum-

venting the regulation, this has (large) negative externalities disregarded by the individual.

Other regulations harm welfare, so corruption limits the harm: It is easy to mention

regulations that mainly serve to produce rents to politically influential groups. This, e.g., applies

to most tariffs. Also, in many less-developed countries it is a problem that the time and effort

needed to obtain legal property rights to a business are far too large; see de Soto (2000).

While such grease-cases exist, they often have dynamic side effects that may change

the conclusion: Perhaps the risk of corruption has made it necessary to have several layers of

expensive controls that slow down the administration, so it is corruption that turns corruption

into grease! A further aspect of the story is that the regulators may slow down administrative

processes precisely to extract bribes. To understand such cases needs a complex model, where

the solution is fragile depending upon the case-dependent details of the model.

Several researchers have tried to estimate reduced form models, with both grease terms

and sand terms. Normally they both become significant, but they are hard to sort out; see Méon

and Sekkat (2005) and Méon and Weil (2010) for somewhat different results.

As the (p2) effect is small, it is possible that it can be turned to become positive by

effective policies to combat corruption. A huge literature discusses such efforts.8

2.3 The internal dynamics of corruption

Many papers, starting perhaps with Andvig and Moene (1990), argue that corruption is dyna-

mic: Corrupt countries tend to become more so, and honest countries become more so as well.

Thus, corruption have a high and a low equilibrium. Paldam (2002) gives an overview of some

mechanisms having this ‘seesaw’ dynamics:

(i) It is impossible to punish everybody if they are all corrupt, but if few are corrupt,

they can be punished. (ii) The corrupt needs to announce his business, and it is typically done

by conspicuous consumption – driving a Mercedes Benz is the classical method in poor

countries. With low corruption such advertisement announces a criminal. (iii) Jobs have

different potential for corruption, and the jobs with the highest potential see wages competed

down, so that the honest seek other jobs. Thus, the corrupt and the honest sort themselves out

in jobs by high and low potential for corruption – this increases corruption.

This suggests that corruption is stuck at rather low T-values in most countries, as seen

below, but once it starts to fall, the rise in the T-values is quite large. This helps explaining why

8. The 1st of February 2020 Google Scholar gave 123,000 hits to ‘anti-corruption policies’.

8

the transition of corruption happens late in the process of development; see section 4.

2.4 Corruption and institutions

The first problem when using institutional explanations is that ‘institutions’ is a wooly term,

and there are many institutions. However, indices for ‘systems’ of institutions have been

compiled. They are multidimensional, so measurement requires indices that cover many aspects

of society. I use the Polity index, P, to catch the political system, and the Fraser index, F, to

catch the economic system, see Table 2. The T-data contain a neat transition curve as shown in

the next section, but so does both the P and the F data; see Figure 6 below. Thus, y, T, P and F

are rather confluent due to joint transitions – relations between these variables include a great

deal of spuriousness.

Corruption and political institutions: Apart from the confluence, it is likely that demo-

cracy and honesty reinforce each other. When civil servants are honest, people are more likely

to trust elections, and hence the elected politicians. Thus, the correlation of the T-series and the

Polity index, P, is likely to be more than spurious. The D-index, defined in a moment, is meant

to eliminate spuriousness.

Corruption and economic institutions: People have different economic ideologies. The

Fraser index is announced as a measure of the freedom to run a private business. Thus, apart

from the confluence, F is linked to corruption in another way: Corruption often occurs to evade

public regulations, and high values of the Fraser index indicate that few such restrictions exist.9

Once again, the D-variable should reduce spuriousness in the relation.

The empirical analysis also uses three measures of institutional instability defined in

Table 2. There is a parallel to the effect of poverty. Uncertainty is hardship that is likely to

increase corruption. Unfortunately, also instability has the problem of confluence, as it

decreases with development; see Paldam and Gundlach (2018).

The empirics of sections 3 and 4 strive to sort out the effect of development as proxied

by income from other effects analyzed.

9. Centrally planned countries have a vast scope for corruption. This helps understanding why such countries have always had large and dangerous elite police corps using wide nets of secret informants.

9

3. The two points (p1) and (p2) and the D-variable

Recall that (p1) is the large positive correlation of T and y, while (p2) is the small negative

correlation between T and g. Section 3.1 estimates the transition path, and calculates the D-

index (already used in Table 1). Section 3.2 looks at correlograms between growth, g, and the

T and D series, while section 3.3 analyzes the inertia in the series.

3.1 The transition path and the D-variable

Figure 1 shows the overlapping data for T and y for the Main sample of 152 non-OPEC countries

(N = 2,730), and in addition Figure 2c covers 15 OPEC countries (N = 247).

Figure 1 shows how the correlation between y and T looks. The kernel regression, T =

K(y, bw), looks as a typical transition curve: It has a flat section at low incomes to the left, and

it is much higher at high incomes to the right, but here the convergence is still going on. The

North Western countries have been wealthy for at least half a century, and the black circles for

these countries have a level well above the transition curve. This suggests that the transition

may end at 9 points. The curve has narrow confidence intervals, so it is well defined.

Figure 1. (T, y)-scatter for Main sample, with the kernel regression K(y, 0.3)

Note: Epanechnikov kernel, bandwidth, bw = 0.3, and 95% confidence intervals, N = 2,730. The most deviating group of countries is the North West countries, which are the black circles at the top. The countries in the group are listed in Tables A1 and A2 (appendix). The vertical dotted line shows that the leftmost 10% of the curve is supported by 1% of the observations only.

10

Figure 2 reports 15 additional kernel curves showing the robustness of the basic path.

Figure 2a shows the robustness of the kernel curve to the bandwidth, bw. As usual, the kernel

is a bit wobbly for small bandwidths and becomes more and more linear (and flat) for large

bandwidths, but the basic form is robust. Figure 2b reports that the curve is stable over time –

though it does move marginally to the right. Figure 2c shows that the transition curve has the

same form in the Main sample and the OPEC countries. As the OPEC countries are relatively

wealthy at each level of development, the K-curve shifts to the right for these countries.

Figure 2. Analyzing the robustness of the K-curve from Figure 1

Figure 2a. Six bandwidths, bw Figure 2b. Five periods

Figure 2c. Main vs OPEC sample Figure 2d. The transition kernel and its reverse

Note: See Table 1. Some estimates use fewer estimates and hence the bw is higher.

Figure 2d is more complex, as it reports an informal causality test. It compares the transition

curve (assuming causality from y to T) with the kernel assuming the reverse causality. Due to

sorting and averaging, the two reverse kernels often look rather different (see Paldam 2019).

11

This is also the case at present. While the T = K(y, 0.3)-kernel generates a beautiful kernel in

accordance with the transition theory, the y = K(T, 0.3)-kernel is difficult to explain, and it has

bends that are not expected from the sand theory.

Thus, K(y) is the transition path for corruption. From that path the D-index is defined

as: Dit = Tit – KT(yit, 0.3). D is negative if the country has ‘too’ much corruption, and it is

positive if the country has ‘too’ little corruption at its level of development. Table 1 showed

that the correlation of D and y are from -0.04 to 0.12. This is insignificant, precisely as it should.

3.2 Properties of the D-series: The deviations from the transition path are rather durable

The D-index shows the deviations from the transition path. The average number of observations

for T and D per country is 18.1. Thus, a t-ratio can be calculated for the D-index for each country

to test if the Ds are above or below the transition curve.

Figure 3 shows the frequency distribution of the 166 t-ratios. The white bars for t’s in

[-2, 2] are the countries that do not deviate significantly from the transition path. The dark gray

bars are thus significant – they are 64% of the countries. Thus, most countries deviate systema-

tically over several decades from the transition path. They are either too corrupt or too honest.

However, even the relative corruption does not explain the growth rates as already shown.

Section 5 gives some examples of such countries, and shows that they are explained by

economic and political instability. Figure 3b looks at the slopes in the T-index for the countries.

Remarkably few of these slopes are significant, which tallies well with high t-ratios of the Ds.

Figure 3. Frequency distribution for the t-ratios and slopes for Ds of 166 countries

Figure 3a. Histogram of t-ratios Figure 3b. Histogram of slopes

Note: Recall that T is corruption and D is the same net of the transition. Empty bars are for insignificant obser-vations, and light gray are for mixed significant and insignificant. The t-ratios are truncated at + 20. At the negative end, the three extreme countries are Taiwan, North Korea and Paraguay. At the positive end, the four truncated are New Zealand, Cape Verde, Sweden and Denmark.

12

3.3 The first contradiction

The introduction stressed the contradiction between (p1) cor(y, T) > 0 and (p2) cor (g, T) < 0.

It was interpreted as a short-run reaction of corruption to changes. Therefore, I look for sign

reversals in the dynamics of cor(g, T). The transition is taken for the strong long-run relation to

which cor(g, T) must adjust. The study of the dynamics of the (g, T) relation has a problem. The

transition gives trends in the variables, but the D-variable is de-trended for these trends.

Figure 4 is the simplest way to look for sign reversals, as it reports the autocorrelation

in the corruption series. The autocorrelations are significantly positive for four years, but they

turn sign and become negative after 6 years with a peak from 10 to 16 years. For lags above 10

the data quickly becomes rather thin, so the peak is not well determined, but is negative for a

substantial period.

Figure 4. Autocorrelation functions with confidence intervals (+ 2 se)

Figure 4a. In corruption index, T Figure 4b. In D-index, T net of transition

Note: Average of results for all 39 countries with full data. The confidence intervals for the two functions overlap for all lags.

Figure 5 develops the correlations (p2) from Table 1 into correlograms. They are

calculated independently for each country, with enough data. The figure shows the averages.

The data quickly becomes thinner as the lags increase. The intersections with the vertical axis

for no leads or lags are the correlations from Table 1 – it is negative for all four curves. Figure

5a contains the trend due to the transition, so it is difficult to interpret. This trend is deleted on

Figure 5b, which uses the D-data.

The left-hand side of the graphs shows the effect of corruption on growth. It is

significant on Figure 5a, but not on Figure 5b. Thus, the negative correlation from corruption

to growth on Figure 5a is due to the long-run trends in the data. The right-hand side of the

graphs shows the effect of growth on corruption. On Figure 5a the negative growth-corruption

correlation vanishes after 6 to 12 years, and on Figure 5b it is even later than after 12 years, but

13

it does disappear. Thus the contradiction between the short- and the long-run is temporary.

Figure 5. Correlogram between growth, g, and corruption, T and D

Figure 5a. Growth and the T-index Figure 5b. Growth and the D-index

Note: The unlagged correlation is made for each country in isolation, so that each point is an average correlation. The black curve is for all countries, while the 20 most developed countries and the OPEC countries are deleted for the gray curve. Countries with less than 5 observations for T are omitted.

3.4 On causality and the quest for long-run factors

It is famously difficult to establish causality, especially in macro relations. However, I have

three indications:

(i) Figure 2d found that the kernel regression for T = K(y) works well, while the kernel

regression y = K(T) works less well. In the same vein, Figure 6 below shows that the Transition

of Corruption happens relatively late in development.

(ii) The asymmetries in correlograms point to Granger-causality, and Figure 5b is very

asymmetric. The asymmetry is to the right, so once the underlying transition trends are deleted

from the data, growth can predict corruption, but not the other way.

(iii) Gundlach and Paldam (2009) report a set of TSIVs (two-stage instrumental variable

regressions), causality tests on cross-country samples for the T and y-variables.10 The instru-

ments are variables for the development potential of countries, much before development

started. The tests show that the main causal direction is from income to corruption. Thus, both

prior research and Figures 2d and 3b argue that the high correlation between income and

corruption is due to a causal relation from income to corruption, .y T⇒

Furthermore, the analysis has shown that the longer run country levels for corruption

have considerable stability. This is also the case for the deviations from the transition trend. It

10. The TSIV-causality test is updated in Paldam (2020), where the main result is confirmed, but some minor simultaneity is also found.

14

must be due to something that changes slowly. Section 4 assumes that this ‘something’ is

institutions.

Another possibility is culture, which is a soft concept where measurement is difficult,

and much is discussed by way of examples. A typical example is the difference in the corruption

in the North West with Anglo-Germanic culture and Southern Europe with Latin-Mediterranean

culture. The paper argues that the difference is caused by the fact that Northern Europe became

wealthy first, but culture may play a role too. It has also been suggested that culture can be

proxied by religion, as analyzed by Paldam (2001), who found that countries with Protestant

Christianity do stick out as relatively honest, but the Protestant countries seem to be the only

ones that differ. Neither Catholics nor Muslims differ when income is controlled for.

4. Institutional explanations, the two points (p3) and (p49

As the reader will recall from the introduction that (p3) was that cor(T, X) > 0, and (p4) was

that cor(T, varX) < 0 for the two system variables X = P, F. Section 4.1 presents the series used.

Section 4.2 reports a set of regressions, while Section 4.3 compares the transitions in the T-

index and the two institutional indices.

4.1 Data for the political and economic systems: Polity and Fraser indices

The Polity index tries to describe the political system on an authoritarian/democratic scale of

integers from -10 to +10. This index gives three variables: Pit itself and the variabilities ViP and

ZiP; see Table 2. The Fraser index of economic freedom tries to characterize the economic

system by the freedom to run a business on a scale from 1 to 10. This index gives two variables,

Fit itself and the variability ViF.11

F started as an annual index in 2000. Consequently, this section works with the

overlapping sample that covers the 17 years from 2000 to 2016, and 144 countries of which 13

are OPEC members. As seen from Table 3, this does not change the correlation of the two

corruption indices (T and D) to income and growth (y and g).

Table 3 provides a quick first view of the relation of T and D to income and the five

institutional variables. The table refers to points (p1) to (p4) in the introduction. The columns

11. Marshall et al. (2016) and the Polity home page explain the P-index. Gwartney et al. (1996, 2018) and the Fraser home page explain the F-index. The P and F indices have a correlation of about 0.45, which is partly explained by confluence due to the transition in the two indices. The two indices aggregate primary series that are almost fully different and strictly confined to either the political or the economic system as claimed. The correlation between the indices is not due to measurement overlapping.

15

for (p1) and (p2) show virtually the same results as in Table 1, so the reduction of the sample

is not important. The correlations (p3) of both T and D to the levels of P and F are positive and

rather large. In particular, note that the correlations cor(T, P) ≈ cor(D, P); thus, the relation of

corruption and democracy is not spurious, while half of cor(T, F) is due to the common

transition. The relations (p4) of both T and D to the three stability measures are all negative and

mostly significant. They are always smaller for D as expected.

Table 3. Correlations of the corruption variables and institutions for country averages

Correlations Level variables Instability variables From introduction (p1) (p3) (p3) (p2) (p4) (p4) (p4) N = 144 y, income P, Polity F, Fraser g, growth VP, var P ZP, no P VF, var F T, corruption 0.75 0.41 0.74 -0.10 -0.48 -0.25 -0.48 D, net of transition -0.05 0.40 0.39 -0.03 -0.17 -0.12 -0.25

Note: The two Vs are the average numerical change, and ZP is the fraction of years without political system.

4.2 The regressions of Table 4

The table has two panels and four parts: Panel 1 to the left analyzes the relations explaining T,

while Panel 2 brings the same relations explaining D to reduce spuriousness.

Parts 1 and 2 analyze the relations of T and D to the levels of y, P and F. The relations

in Part 2 of the table have strong collinearity, especially of y and F. Regression (T1) explains T

by income, and (T3) explains T by the two institutional variables P and F. Both regressions

have an R2 score of about 0.55, but (T2) shows that when all three variables are in together, R2

only increases by 0.13 to 0.68. When both the corruption transition and income are taken out

of the regression in (D2), the two institutional variables explain 0.21 of the variation. In

regression (D3) the effects (beta) of P and F are the same. Thus, the transition is the strongest

factor explaining corruption, but both P and F have an independent role as well.

Part 3 reports that all three measures of instability have a negative effect on honesty,

especially the two Vs that are almost equally strong. Thus, instability increases corruption. This

tallies well with the strong effect of poverty. Poverty and uncertainty increase corruption, and

it does not matter if the uncertainty is in the economic or political sphere.

Part 4 combines the level and instability variables. While P and VP produce the same

coefficient whether or not the other variable is included, this is not the case for F and VF, where

F knocks out VF. Thus, perhaps the large effect of VF in Part 3 of the table is an artifact. The

table certainly tells a story of collinearity. I return to this knock-out point in a moment.

16

Table 4. Seven OLS regressions explaining T, corruption and D corruption net of transition

Table 4a. Panel 1 explaining T Table 4b. Panel 2 explaining D

Part 1 (T1) (D1) Level Coef t-ratio Beta Same regressions Coef t-ratio Beta Same regressions Income 1.24 (13.4) 0.74 without income -0.04 (0.5) -0.05 without income Constant -6.95 (-8.4) 0.20 (0.3) R2 0.56 0.00 Part 2 (T2) (T3) (D2) (D3) Level Coef t-ratio Beta Coef t-ratio beta Coef t-ratio Beta Coef t-ratio beta Income 0.78 (7.7) 0.47 -0.45 (-5.4) -0.48 P, Polity 0.04 (2.4) 0.13 0.04 (1.7) 0.10 0.05 (3.5) 0.26 0.06 (3.4) 0.28 F, Fraser 0.85 (5.7) 0.38 1.56 (11.0) 0.69 0.73 (6.0) 0.58 0.33 (3.1) 0.26 Constant -8.78 (-11) -6.43 (-7.0) -1.29 (-1.9) -2.63 (-3.8) R2 0.68 0.55 0.35 0.21 Part 3 (T4) (T5) (D4) (D5) Changes Coef t-ratio Beta Coef t-ratio beta Coef t-ratio Beta Coef t-ratio beta Income 0.92 (8.2) 0.55 -0.35 (-3.8) -0.38 VP, var P -1.18 (-3.6) -0.21 -2.26 (-6.2) -0.40 -0.81 (-3.0) -0.25 -0.40 (-1.5) -0.12 ZP, no P -2.42 (-1.4) -0.08 -6.01 (-3.0) -0.19 -3.19 (-2.2) -0.18 -1.81 (-1.3) -0.10 VF, var F -6.11 (-3.1) -0.19 -13.59 (-6.4) -0.41 -7.02 (-4.3) -0.38 -4.15 (-2.7) -0.22 Constant -2.72 (-2.3) 6.99 21.7 4.30 (4.3) 0.56 (2.5) R2 0.62 0.43 0.17 0.09 Part 4 (T6) (T7) (D6) (D7) Both Coef t-ratio Beta Coef t-ratio beta Coef t-ratio Beta Coef t-ratio beta Income 0.62 (5.8) 0.37 -0.60 (-7.0) -0.65 P, Polity 0.05 (3.0) 0.15 0.05 (2.6) 0.15 0.06 (4.0) 0.29 0.06 (3.5) 0.30 F, Fraser 0.80 (5.0) 0.36 1.19 (7.4) 0.53 0.64 (5.0) 0.51 0.26 (1.9) 0.21 VP, var P -1.16 (-4.0) -0.20 -1.68 (-5.5) -0.29 -0.83 (-3.5) -0.26 -0.32 (-1.2) -0.10 ZP, no P -2.30 (-1.5) -0.07 -4.35 (-2.7) -0.14 -3.02 (-2.5) -0.17 -1.03 (-0.8) -0.06 VF, var F -0.18 (-0.1) -0.01 -1.96 (-0.9) -0.06 -1.97 (-1.3) -0.11 -0.23 (-0.1) -0.01 Constant -6.46 (-5.0) -3.03 (-2.4) 1.32 (1.3) -2.03 (-1.9) R2 0.72 0.65 0.43 0.23 Note: For national averages N = 144. The explanatory variables have different scales. To make the effects comparable, the standard estimates are supplemented with beta coefficients, for normalized series.

The positive effect on corruption of the two institutional variables and the negative

effect of the instability of the same two institutional variables are easy to interpret. A transition

is a process of system change. The changes are mainly to the better, so in the longer run

corruption decreases. However, the changes are also instability, so in the short run they increase

corruption; see Paldam (2019). This is a story of short-run costs versus long-run gains. It points

back to (p1) the positive effect of income and (p2) the negative effect of growth from section

17

3. If part of the transition is caused by the transitions in the institutional variables, it reinforces

the idea that the main direction of causality is from the transition (income) to corruption, not

the other way around.

4.3 Comparing transitions: An additional causality indication

On Figure 1 the old wealthy countries of the North West stuck out as unusually honest. This

was interpreted as an effect of time on the internal dynamics of corruption as discussed in

section 2.3. Once countries become honest, they get gradually more honest over time.

Figure 6 is another way to illustrate the late transition of corruption. The figure

compares three transitions estimated on overlapping data for 2000-16: KT estimates the

Transition of Corruption (analyzed above). The curve is almost the same as on Figure 1. KP

estimates the Democratic Transition (analyzed in Paldam and Gundlach 2018). KF estimates

the Transition to a Market Economy (analyzed in Bjørnskov and Paldam 2012, and Paldam

2020). The dotted vertical line at y = 6.6 shows that the low ends of the curves are fragile.

Figure 6. Comparing three transition estimates: KT, KP and KF

Note: Estimated for N = 1,965 for the years 2000-16. The data for y < 6.6 are thin. F is the Fraser index for economic freedom, P is the Polity2 index for the degree of democracy, and T is corruption as on Figure 1. OPEC countries are excluded. The confidence intervals for KP and KT overlap below 8.2.

18

When the KT-curve is compared to the KP-curve of the democratic transition, the paths

of the two transition curves look rather similar, but the KT-curve is one full log-point of income

later than the KP-curve. This is a difference in GDP per capita of 2.7 times, which is growth in

30-50 years. Thus, first the political system becomes more democratic, and after several decades

corruption falls. I interpret this as evidence on long-run causality from P to T.

The transition curve KF is less clear. The curve has a positive slope throughout, but no

signs of convergence at either end; but the confidence intervals (not shown) are narrow, so it

does represent a systematic change. The KF-curve looks as a typical (log-linear) income curve.

This explains the ‘knock-out’ point from the previous section. However, given that the F-curve

is interpreted as a transition curve, it is clear that it starts to rise well before the KT-curve. Thus,

it can explain the KT-curve, and support the conclusion that the corruption transition is late and

due to transitions in other variables, notably institutions.

5 Six examples

The examples illustrate the relations between institutional uncertainty and corruption. The first

three are for country pairs on three continents, where each pair has many similarities and a clear

difference in the level of corruption. Table 5 gives the data for the country pairs shown in

Figures 7 to 9, while Figure 10 looks at three countries with spectacular crises. The four figures

are drawn for the D-data, so the transition curve is horizontal at D = 0 per definition.

Table 5. The institutional variables in 3 country pairs Argentina Chile Latvia Estonia Côte d’Ivoire Ghana P 8.12 9.65 8.00 9.00 1.88 7.29 ZP 0 0 0 0 0.16 0.02 VP 1.00 0.52 0.43 0.46 0.27 1.12 F 5.78 7.75 7.55 7.80 5.73 6.53 VF 0.26 0.07 0.13 0.08 0.11 0.15 Start 1995 1995 1998 1998 1998 1998

Note: The period always ends in 2017. The values given are the averages for the periods given. The table gives the most corrupt of the pair first and shaded.

5.1 Pair 1: Argentina and Chile

The two neighboring countries on the Southern Cone are both Spanish ex-colonies, with much

the same immigration history, language, religion, etc. They are also at the same income level,

though Chile has grown much faster. Still, the level of corruption differs by 4.2 points.

19

Figure 7. D-levels over income in Argentina and Chile, 1995-2017

The two institutional indices show that Argentina has had less democracy and economic

freedom. Argentina has also had much more volatility both in the political and economic system

– and the differences started long before the indices. Thus, it fits our story perfectly well. It is

difficult to explain why the two countries had such a different history, but once things started

going awry in Argentina, there was an amazing lack of brakes

5.2 Pair 2. Estonia and Latvia

The two Baltic countries Estonia and Latvia have had much the same history – at least since

1795 where they both were integrated into Russia as provinces. They were independent from

1918 to 1940, where first a brief Russian and then a German occupation took place. Finally,

they returned to Russian rule until liberation in 1990/91.

Figure 8. D-levels over income in Estonia and Latvia, 1998-2017

20

In spite of this common history, and a similar income level and population size, the

two countries have a difference of 1.6 points in the level of corruption. The level of the

institutional variables are higher in Estonia, while there is no difference in the instability

variables, but perhaps it is not so much the actual instability that counts as the potential one:

Latvia is a much more divided country, both as regards ethnicity and religion. This surely

gives some uncertainty.

5.3 Pair 3: Côte d’Ivoire and Ghana

Côte d’Ivoire and Ghana are (also) neighbors, at roughly the same size and income level, but

they differ as to colonial history, languages and ethnicity. The T-index differs by 1.3 points.

As regards the institutional variables there has been a dramatic shift: In the 1970s Ghana

was a much more unstable country, but now it is the other way round. The level of the

institutional indices are (much) higher in Ghana, while the volatility variables give a more

unclear picture. The ZP-variable is 0.26 in Côte d’Ivoire. It reflects that the country has had

about 5 years of civil war in the period. That the difference in the level of corruption is not

larger is probably due to the previous period, where Ghana fared badly. Note also that the two

low-income African countries have had a more volatile economic development than the middle

income countries of the previous pairs.

Figure 9. D-levels over income in Côte D’Ivoire and Ghana, 1998-2017

5.4 Three crises: Greece, Venezuela and Zimbabwe

Figure 10 shows the development in the D-index in three countries (on different continents)

that have experienced very big economic crises: Greece, Venezuela and Zimbabwe. Their D-

21

indices are depicted with time on the horizontal axis. Note that as the D-index is used, the curves

do not reflect differences in corruption, but in corruption relative to other countries at the same

income level.

The crises in the three countries were preceded by at least a decade of inconsistent

economic policies that sober observers soon found irresponsible. At some stage the policies

caused galloping debt, balance-of-payment deficit and increasing inflation, and finally a large

fall in the gdp. This sequence led to a fall in the D-index in all three cases, where the index

turned negative. The small vertical lines indicate main events.

Greece became relatively corrupt around 2000, maybe as the Greeks became cynical as

regards the policies pursued, and it increased quite strongly, but temporarily only, during the

full scale crisis 2008-13. However, Greece remains a relatively corrupt country.

Figure 10. The story of the D-index in three crisis-countries

Venezuela has fared rather poorly for a long time, in spite (or because) of its oil wealth.

This led to the political victory of the populist Hugo Chaves, who was president 1999-2013.

Maybe 2002 was the turning year where his policies became unsustainable. The economic

balance in Venezuela gradually worsened, and corruption that was already too high increased

by further two points. After the death of Chaves, power went to his vice-president Nicolás

Maduro, who continued his policies with catastrophic results. The upturn in the last two years

of the D-index is due to the large fall in the income level. The T-index remains constant at 1.7-

1.8, making Venezuela the most corrupt country in Latin America.

Zimbabwe was known as a relatively honest country till the rapid socialization program

22

was started in 2000, but then corruption increased by about 2 points. During the dramatic

debacle of the economy, corruption remained rather trendless.

When this evidence is summarized, it is clear that an economic crisis increases

corruption, but the timing is not so clear. It probably depends upon the extent to which people

understand what is going on.

6. Conclusions

The paper has shown a strong transition in the level of corruption as measured by Transparency

International’s T-index. Poor countries are rather corrupt, but they become honest as they grow

wealthy. The transition is a complex process that interacts with institutions. They also have

transitions, so the relations examined contain a great deal of collinearity. When the transition

path is deducted from the T-index – to give the D-index – it greatly reduces multicollinearity,

and allows an identification of the substantial genuine effect of institutions.

The transition of corruption happens relatively late in the development process. The

lateness argues that the transition, and hence T, is caused by development and not the other way

around. This tallies with the long-run causality test in Gundlach and Paldam (2009).

A main reason for the late transition is that development creates many changes that

inevitably give uncertainty, which causes setbacks in the corruption index. Such setbacks are

temporary, and when institutions stabilize and countries become stable, wealthy, liberal

democracies, honesty comes to dominate. Thanks to the short-run reversals and the internal

dynamics of corruption, the process takes time.

23

References:

Aidt, T.S., 2003. Economic Analysis of Corruption: A Survey. Economic Journal 113 (491), 632-52. Also chpt.

3 in Dutta and Aidt (2016)

Andvig, J.C., Moene, K.O., 1990. How corruption corrupt. Journal of Economic Behavior and Organization 13,

63-76. Also chpt. 16 in Dutta and Aidt (2016)

Bjørnskov, C., Paldam, M., 2012. The spirits of capitalism and socialism. A cross-country study of ideology.

Public Choice 150, 469-98. See also update in Paldam (2020) on the F-index.

Chenery, H.B., Srinivasan, T.N., eds., 1988, 1989. Handbook of Development Economics, vol. 1 and 2. North-

Holland, Amsterdam

Dutta, J., Aidt T.S. eds., 2016. Corruption and Economic Development. The International Library of Critical

Writings in Economics series, vol. 324, Elgar, Cheltenham

Fraser index of economic Freedom. URL: https://www.fraserinstitute.org/studies/economic-freedom. See also

Gwartney et al. (2018)

Gründler, K., Potrafke, N., 2019. Corruption and economic growth: New empirical evidence. European Journal

of Political Economy online first

Gundlach, E., Paldam, M., 2009. The transition of corruption: From poverty to honesty. Economic Letters 103,

146-48. Also chpt. 27 in Dutta and Aidt (2016)

Gutman, J., Paldam, M., 2020. Comparing corruption data. Forthcoming

Gwartney, J., Lawson, R., Block, W., 1996. Economic Freedom of the World: 1975-1995. Fraser Institute,

Vancouver, B.C. First annual publication of the Fraser index.

Gwartney, J., Lawson, R., Hall, J., Ryan Murphy, R., 2018. Economic Freedom of the World 2018 Annual Report.

Fraser Institute, Vancouver, B.C. See also Fraser index for home page.

Heidenheimer, A.J., Johnson, M., LeVine, V., eds., 1989. Political Corruption: A Handbook. Transaction

Publishers, New Brunswick, NJ.

Lambsdorff, J.G., 2007. Institutional economics of corruption and reform. Theory, evidence and policy.

Cambridge UP, Cambridge UK

Leff, N. H., 1964. Economic development through bureaucratic corruption. American Behavioral Scientist 8, 8-

14, and in Heidenheimer et al. (1989), chpt. 24 pp. 389-403, and chpt. 23 pp. 553-59 in Dutta and Aidt

(2016)

Marshall, M.G., Gurr, T.R., Jaggers, K., 2016. Polity IV project. Users’ manual. University Center for Systemic

Peace. See Polity index for home page

Méon, P.-G., Sekkat, K., 2005. Does corruption grease or sand the wheels of growth? Public Choice 122, 69–97

Méon, P.-G., Weil, L., 2010. Is Corruption an Efficient Grease? World Development 38, 244-59. Also chpt. 24 in

Dutta and Aidt (2016)

Net appendix is available from http://martin.paldam.dk/Papers/GT-Main/T-appendix.pdf

Paldam, M., 2001. Corruption and religion. Adding to the economic model. Kyklos 54, 383-414

Paldam, M., 2002. The big pattern of corruption. Economics, culture and the seesaw dynamics. European Journal

of Political Economy 18, 215-40. Also chpt. 25 in Dutta and Aidt (2016)

Paldam, M., 2009. The macro perspective on generalized trust, chpt. 21, pp. 354-78 in Svendsen and Svendsen

24

(2009)

Paldam, M., 2020. The Grand Transition. Book manuscript updating and expanding the papers in my project with

Erich Gundlach. Available from http://martin.paldam.dk/GT-Book.php

Paldam, M., Gundlach, E., 2018. Jumps into Democracy: Integrating the Short and Long Run in the Democratic

Transition. Kyklos 71, 456-81

Polity index. URL: http://www.systemicpeace.org/polityproject.html. See also Marshall et al., 2016.

Soto, H. de, 2000. The Mystery of Capital. Why capitalism triumphs in the West and fails everywhere else. Black

Swan Books, London

Svendsen, G.T., Svendsen, G.L.H., eds., 2009. Handbook of Social Capital. The Troika of Sociology, Political

science and Economics. Elgar, Cheltenham

Treisman, D., 2000. The causes of corruption: a cross-national study. Journal of Public Economics 76, 399-457

Ugur, M., 2014. Corruption’s direct effect on per capita income growth: a meta-analysis. Journal of Economic

Surveys 28, 472-490