1) What it Takes to Shape Economic Growth? or 2) Assessing the Impact of Political Leaders on Economic Growth

I first revisited Jones and Olken (2005) methodology by using only leaders who died in power by natural death. Thus, with more recent data I[r]

1) What it Takes to Shape Economic Growth?

or 2)Assessing the Impact of Political Leaders on

Economic Growth

Julieta Peveri May 2019

AbstractThis paper analyses the context in which national leaders can shape economicgrowth and it assesses which individual characteristics are linked to their perfor-mance By comparing growth variations across transitions in which the identity

of the ruler is not likely to be dependent on the economic conditions, I found idence that leaders matter both in autocracies and democracies even though thelevel of democracy and development shrink the magnitude of leaders’ effects Be-sides, rulers’ transitions have a higher impact on growth whenever one of the leadershas the possibility of running for a second mandate In some specifications, I alsofound evidence of the positive effect of university education and a mitigate effect

ev-of tenure on growth shifts Regarding directional results (i.e nominal growth tions) individual traits play a larger role Younger leaders and the ones with moreexperience in politics are associated with a better economic performance Somecareer backgrounds are also relevant for establishing the leader’s quality as well

varia-as the existence of a term limit In some of the approaches used, more educatedleaders appear to perform better

1 Introduction

At the intersection between Politics and Economics the leader’s identity is becoming atopic of increasing interest But can they matter to the point of shaping economic growth?From a theoretical classic framework the answer would be negative Indeed, most of theacademic research has highlighted, as determinants of growth, factors such as capitalaccumulation, innovation, geography or education Hence, while those characteristicshave been stable over time, growth rates have not, particularly in developing countries(Pritchett, 2000) Alternative explanations have tried to understand the growth ratesvolatility For instance, Easterly et al (1993) show the importance of shocks, particularlythose in terms of trade as a determinant of the long run growth variance As thosevariations are random, they conclude that high growth rates are likely due to good luck

In that sense, they show a similarity with the classic explanations by delegating the blame

of underdevelopment or the merit of economic miracles to factors beyond our control

In this paper I focus on the potential role of political leaders to account for growthinstability The challenge of identifying a causal impact is that growth can also triggerleaders transitions In democracies, reelections can be tied to the previous economicperformance and in autocracies, coups are less likely to arise during economic booms(Londregan and Poole, 1990) Jones and Olken (2005) tackle this endogeneity issue byfocusing on transitions where leaders died in office by natural death, considering those asunpredictable events Indeed, they found positive evidence suggesting that rulers matterfor growth but only under autocracies The first contribution of this paper is to revisittheir work by using more recent data Thereafter, I find positive evidence for the effect

of leaders also under democracies Hence, this strategy is mainly limited by the number

of observations and the lack of representatives of the ones who died in office

In order to go deeper in the analysis, a larger number of exogenous transitions isneeded For this purpose, I also include the ones where the exiting leader was not able

to run for a reelection due to a term limit constraint Besides, I select transitions wherethe entering ruler either won the elections by a small margin of victory, took powerthrough royal succession, by constitutional order, was elected by an elite or as an interimleader (as long as the predecessor exited in a regular manner) Using those four hundredcases, it is possible to give a brighter insight of the context in which leaders are morelikely to have a stronger impact on growth by including further contextual variables andleaders’ characteristics The evidence suggests that leaders have more power in low-income countries and in more autocratic ones, being less constrained by the institutions.The possibility of reelection also accentuates the growth volatility Overall, leaders’ traits

do not play a role to determine the amplitude of their effect in the economy

The other main question addressed in this paper is whether there are observableindividual attributes that, given a specific context, can help us differentiate between good

and bad leaders While at the local level, many studies have given attention to politicians’characteristics such as gender, education and age, at the national level, the empiricalevidence is limited For instance, Besley et al (2011) extended the analysis of Jones andOlken (2005) and tested if the impact of leaders (also in transitions where the ruler died

by natural death while in office) varied according to his educational attainment Again,the analysis was limited to the number of observations Here, I explore and interactfurther individual features and control for contextual variables While I found that therobustness of education is not strong (even though the signs are always consistent with theprevious findings), the results suggest that young leaders and the ones with more years

of experience in politics are associated to a better performance Lawyers and militaryleaders also appear to boost economic growth

The paper is organized as follows Section 2 reviews the existing related literature;Section 3 describes the data and sources used in this work; Section 4 explains the method-ology used in this paper; Section 5.1 revisits Jones and Olken (2005) approach focused onthe rulers who died in office by natural death, Section 5.2 includes new cases of exogenoustransitions; Section 6 provides some robustness about the validity of the methodology andalternative specifications for growth comparison and Section 7 sets out a conclusion

There is no consensus in theory about the significance of political leadership Even inthe historical literature the opinion is divided Defenders of the Great Man theory such

as Thomas Carlyle or Sidney Hook argue that history can be largely explained by sions and actions impulsed by certain leaders On the other extreme, many philosophersconsider historical events and leaders as determined by the society it self Spencer (1892)states: “You must admit that the genesis of a great man depends on the long series ofcomplex influences which has produced the race in which he appears, and the social stateinto which that race has slowly grown Before he can remake his society, his societymust make him.”

deci-When it comes to the Political Economy literature, the ambiguity about rulers’ tity is also there In the branch of political competition, models that rely on the medianvoter theorem predict a convergence of all candidates on one policy platform, while othersthat account for parties and politicians ideologies lead to an equilibrium that is linked tothe identity of the political leader

iden-Another concept of great interest for this work is the role of institutions that has beenhighlighted by several authors such as North (1992); Acemoglu and Robinson (2012);Acemoglu et al (2001); Rodrik et al (2004); Acemoglu and Johnson (2007); Gaeser et al

(2004) Even by agreeing on the importance of institutions, the role of the politicalrulers is not clear It can be argued that when institutions are strong they would act

as a constraint on the incumbent leader In this sense, rulers shouldn’t have an impact

on consolidated democracies but they might so in autocracies For instance, Acemoglu

et al (2003) show that once controlling for institutions, macroeconomic policies have asmall impact on growth volatility Nevertheless, being the nation’s leader the main agentamong political institutions, they could have through this power an indirect impact oneconomic and social outputs in any regime

Even though we reach a consensus that politicians can matter for some economic andsocial outputs, it is still a strong statement to argue that they can shape the growthpattern To the best of my knowledge, the only study to assess national leaders as

an alternative explanation for the fluctuating path of growth is the one by Jones andOlken (2005) They explore whether the changes of national leaders could be related

to the variability associated with growth rates Their empirical strategy was to exploittransitions in which the leader died by natural death while in office, considering those

as “random” ones The authors tested whether the growth pattern before the leader’sdeath was statistically different after his replacement After finding a positive answer

in an autocratic context, they reproduced the analysis about the effects of leaders onparticular types of policies outcomes and they found a significant one on monetary policybut ambiguous evidence for changes in physical and trade policy There is also evidenceprovided by Fatas and Mihov (2013) that policy volatility can have a strong and negativeimpact on growth

If identity appears to matter, it is natural to ask which of the individuals istics lead the magnitude and the direction of the effect Following the work of Jonesand Olken (2005) cited before, Besley et al (2011) extended the analysis and tested ifthe impact of a leader varied according to his educational attainment They also found apositive answer, consistent with the findings related to the impact of education on earn-ings in labor economics This branch of the literature states that education is seen notonly as the reflect of specific knowledge and skills taught in the academic space but also

character-as a proxy to unobserved ability In that sense, firms are willing to pay higher salaries

to more educated workers and more experienced ones because they associate those viduals with a better performance By extrapolating the argument for a nation’s head,one may expect that the more skilled, competent and qualified the leader is for his po-sition, the better will be the decisions he will lead to, thanks to his expertise, academicknowledge and the unobserved ability that encompasses motivation, intelligence, facility

indi-to learn, capacity of solving problems, among others

At the local level, studies that assess the characteristics of leaders on policy outcomesare more frequent as it is easier to design an empirical strategy to identify a causaleffect For instance, through a regression discontinuity design Lahoti and Sahoo (2017)exploit data from Indian politicians, finding that educated leaders were not necessarilymore competent in improving children’s education Controversely, Diaz-Serrano and Prez(2013) study concludes that the educational attainment of population does improve when

a leader with higher education remains in office Besides education, the impact of theleader’s gender has also become an active research topic Chattopadhyay and Duflo (2004)exploit data from India where some council head positions have been randomly allocated

to a woman They find that leaders spend more in infrastructure that is directly relevant

to the needs of citizens of their same gender Brollo and Troiano (2016) use a discontinuityapproach based on close elections and found that in Brazil female mayors are less likely toengage in corruption, to hire fewer temporary public employees during the electoral yearand they also tend to attract less campaign contributions when running for reelection.Using the same empirical strategy, Alesina et al (2015) analyze the effect of leaders’ age

on political governance, reelection rates and policies in Italian municipalities

Horowitz and Stam (2014) who study how leaders’ characteristics affect military cisions, provide an intuition illustrated by Figure 1 of how leaders’ experience may affectpolicies choices

de-Figure 1: Theoretical relationship between leader’s experiences and policy outcomes

Life experiences Leader beliefs/

risk attitudes Policy outcomes

Domestic politics

Source: Horowitz and Stam (2014)

The managerial literature can also provide valuable analysis through the importancethey have given to the impact of the CEO on the performance of a firm by analyzing theirleadership style, their risk-taking behavior and some personal traits such as age, gender

or family social class One specific paper related to the research question stated above isthe one written by Bertrand and Schoar (2003) that analyzes the effect of the CEO on theperformance of a firm through estimating their fixed-effects Their study concludes thatdifferences in investment, financing and other organizational strategy variables depend

on the specific characteristics of the firm’s manager

Another interesting link can be done on the Political Psychology field, in which

schol-ars have analyzed the behaviour of leaders since long time ago The main approach used

is the psychobiographic method which involves an examination of the life history of rulers.Many of the personality theorists seek to identify traits (defined as personality character-istics that are stable over time and different situations) motivations and cognitive stylevariables to assess how all these features shape styles of decision making, interpersonalinteraction and management in office (Cottam, 2004) For instance, Adorno et al (1950)(cited in Cottam (2004)) study the personality associated to authoritarian leaders show-ing that it was a result of childhood experiences that led to a weak ego Regardingdemocratic chiefs of state, Dean Keith (1993) analyzes the way we judge who would bethe best leader for a nation among a pool of candidates As the author states: “Somecharacter traits go better with certain policy stands or performance expectations Forinstance, our assessments of a candidate’s willingness to solve the problem of the home-less may depend in part on our perceptions of how compassionate we perceive him or her

to be Similarly, we may feel that certain personality traits may enhance a candidate’sprospects for effective performance” (Dean Keith, 1993)

To begin with the analysis of the relationship between political transitions and growth,Figure 2 plot the evolution of the real growth trend for four selected countries where thevertical lines represent the transitions of chiefs of state At first sight, it is possible todetect that a great proportion of those political changes lead to non negligible shifts inthe real GDP Besides, contrary to what we could expect in many of those transitionsthe variations on the economic output are manifested immediately

One of the main reasons of the lack of empirical research about national leaders in theeconomic field may be related to data limitation Fortunately, in recent years progresshas been made In 2004, the Archigos dataset (Goemans et al., 2009) with informationabout the year and the nature of entry and exit of national leaders in 188 countries from

1875 to 2004 was published It also contains few other personal variables such as gender,the year of birth and the year and cause of death From then on, other contributionshave been made by completing this set with further information This paper combinesthe LEAD base (Elli et al., 2015) with the Cursus Honorum (Baturo, 2016) one to obtain

a large number of information related to the leader’s traits and background

The Leader Experience and Attribute Descriptions (LEAD) dataset (Elli et al., 2015)covers information about 2 964 national leaders from 1840 to 2000 It is based on theArchigos dataset but it includes additional variables related to family background, profes-sional and personal history, health status, education, military experience, among others.Their creators are using this dataset for international conflict studies The Cursus Hon-orum Dataset (Baturo, 2016), also based on the Archigos one, includes 1 500 political

Figure 2: Real growth path and political transitions in selected countries

Source: Bolt et al (2018)

leaders in office from the period between 1960 and 2010 As the LEAD data base, italso includes educational and career variables as well as other ones related to the familybackground Unfortunately, many of those variables have a lot of missing values, and asmuch I think they may affect the quality of a leader, I do not consider them

Gender is one of basic traits defining an individual However, in the remaining sections

I do not analyze the effect of this characteristic do to the few national female leaders

In fact, men represent an extremely high percentage of 98% of the dataset (see Table1) Even if one might think that this proportion is decreasing nowadays, if we focus

on the sub-sample from 1990 to 2004, it only decreased to 95,5% and even goes up to96,7% when we consider the period 2000-2004 In his book, Ludwig (2002) argues thatpeople associates authority with masculinity traits More interesting, in his analysis(previous to 2002), he shows how almost half of the women that have been head of stategained this position for having been the widows or daughters of previous leaders Amongthe remaining women rulers that become leaders for her own merits, most of them juststayed less than a year in power The author also points out that no woman has lead an

Table 1: Descriptive Variables

be arguably regarded and seen as an outsider to the isting political system Political family is coded 1 if a member of the leader’s family had occupied the highest national political posts in the past, whenever possible

ex-to ascertain.

autocratic regime in the history of the world

Another basic variable to analyze is age When plotting the mean age when leadersentry office in Figure 3 we can see almost no variation from the 90’s on, being the meanage 54.4 years old (in this period) with a standard deviation of 2.14 years old Thus, wecan argue that young people do not have more access to the head of nation’s positionthan before as we could presume

Concerning some family variables, it appears that the majority (55%) of nationalleaders (for whom this data is available) come from middle class families while a 24%have grown in a low class family and the remaining 21% were raised in families with a highsocio-economic level Particularly relevant for the forward analysis are the educationaland professional variables Figure 4 illustrates the percentage of the overall level ofleaders’ education by decade It is interesting to see how in the last two decades ofanalysis the proportion of leaders with a postgraduate education has increased and less

Figure 3: Mean age of leaders when entry into office by year

Source: Elli et al (2015)Figure 4: Leaders’ level of education by decade

Source: Elli et al (2015)

than 15% of leaders have not gone to University over the last decade

Then, Table 2 highlights that despite chiefs of state come from all kinds of grounds, the ones from the political and legal fields are clearly a majority Indeed, the

back-Table 2: Leaders’ Profession

Source: Elli et al (2015)

percentage of national leaders with a military career is also high It is also helpful toanalyze the nature of leaders’ exit shown in Table 6 In Section 5.1 I will exploit data

of leaders who died in office which represent around 8% of the dataset As we couldexpect, leaders who died in office are in average older when taking power (63 years old)and around half of them had run an autocratic government

Table 3: Leaders’ Exit Type

Source: Goemans et al (2009)

The data for economic growth used in this dissertation is drawn from different versions

of the Pen World Table (Heston et al., 2002, 2009; Feenstra et al., 2015) and the MaddisonProject (Bolt et al., 2018) Figure 5 shows that it exists a positive correlation between theaverage growth rate and the leader’s educational level until university studies Yet, the

growth rate associated to having a postgraduate education is not significantly differentfrom the one associated with a university diploma This paper also uses data from Beck

et al (2001) and Ginsburg et al (2010) to identify mandates where the leader have ornot a term limit constraint

Figure 5: Correlation between growth and leaders’ educational level

Source: Elli et al (2015) and Heston et al (2002)

In the first part of this paper I will replicate the methodology proposed by Jones andOlken (2005) but using alternative data sets both for growth and leaders The startingidea is that the growth process can be described by Equation 11

where git is the growth rate of country i in period t; vi represents a time fixed effect; lit

is an approximation to the leader’s quality and it is the error term which is assumed tohave mean 0 and variance σi2 Just like the authors, I assume region-specific heteroskedas-ticity and a region-specific autocorrelation process of first order The hypothesis to betested is whether λ is different from 0 meaning that political leaders can contribute toshape economic growth

Hence, the main problem of interpreting directly the coefficient of leaders fixed-effects

is that leaders transitions are not likely to be exogenous to economic conditions Asexplained by Jones and Olken (2005), this can arise if the probability that a leader isselected in a country i at a time t depends, among other variables, on the previous growthpath Formally:

where l’ is distributed normal, with mean µ variance σ2

l and corr(l, l’) = ρ This reflectsthe main identification problem when it comes to asses the impact of a leader on growth.For instance, in democracies where the incumbent president can run for reelection, voterswould, among other things, take into account the economic performance during his term

to either replace or reelect him Also, in autocracies the probability of revolution torevoke the dictator is likely to depend on the level of inequality and living standards

To deal with the endogeneity issue, Jones and Olken (2005) suggest to use leaderswho died in office by natural death so the related date of the transition was not predictedand assumed to be independent to economic conditions Rather than focusing on leadersfixed effects, Jones and Olken (2005) choose to compare differences in the average growth

T periods before the death of a leader and T periods after it More specifically, being

\

P REz the average growth in the T years before the leader l dies and \P OSTz the average

in the T years that follow the death of the leader, if the growth process is described

by Equation 11, then the difference between \P REz and \P OSTz will be distributed asfollows:

Θj = 1

(a) if lij−1 died by natural death

(b) if lij−1 could not run for elections due to a constitutional term limit and

lij−1 exited in a regular manner

(c) if lij won the elections by less than 55% of the votes and lij−1 exited in aregular manner

(d) if lij assumed through royal succession, constitutional succession, elected

by an elite or as an interim leader and lij−1 exited in a regular manner

where lij represents the jth leader of country i

The first case was analyzed above In the second case, as the incumbent leader isunable to run for elections, inevitably a new leader has to take power independently

of the previous growth pattern Nevertheless, some endogeneity can still be present ifcandidates are judged based on the performance of leaders from their same party In thethird case I assume that if a leader takes power following a close election, his victory can

be considered as random Finally, I add transitions in which the incoming leader was aninterim one; he assumed power by a royal succession; a constitutional one or if he wasselected by an elite I assume that in those cases the motivations of changing the leaderwere not based on the previous growth path I eliminate cases in which, even though one

of the previous conditions held, the leader who exited power did it through an irregularmanner (such as coup, revolution or murder)

Just like before, I first tested whether growth rates, N years before lt−1 leaves theoffice, are in average significantly different N years after he left with the same Waldtests previously used Hence, in addition to confirming that rulers do have an impact ongrowth, I also wish to study in which context they are more likely to shape growth, Iprivilege another approach by estimating the following equation:

|∆growth\j−1,j| = α + β∆Xj,j−1+ φ1Yj + φ2∆Yj−1+ λRegion+  if Θ(lj) = 1 (6)

where j∆growth\j−1,jj is the absolute value of the average growth difference betweenthe growth rates in the first three years of leader j (or less if the leader stayed one or twoyears) and the previous three (or less) years of his predecessor;1 ∆Xj,j−1is a vector of thedifferences of individual leaders or country’s continuous characteristics; Yj and Yj−1 arevectors of leader or country dummy’s variable corresponding respectively to j’s and j-1’sterm and λRegion are region fixed-effects Finally, I also use the nominal values of thosedifferences in growth rates to assess the quality of the leader rather than the magnitude

of his impact in order to assess if there are some observable characteristics that may allow

1 By not considering the whole term, we mitigate the problem about the correlation between the end dates of leader’s j and the economic conditions.

us to differentiate the rulers that manage to boost economic growth.

In addition to the endogeneity of transitions, there is also the concern that the growthrate of the first year of each term is likely to be constrained by decisions made by theprevious leader and the last years may not be representative of his average performance

if the motivations are different (specially in transitions with term limits) I suggest tosmooth growth rates as follows:

This approach slightly moderates the first and last year Indeed, this seems ate as I already assigned the transitional year to the ruler who stayed longer in power.Thus, during the in and out year, each leader stayed more than six months Nevertheless,

appropri-in Section 6 I explore other weightappropri-ing coefficients

I begin by using the same methodology and the same growth rate data as Jones andOlken (2005) (i.e the real GDP per capita growth rate at constant prices using Laspeyresindex obtained from the Pen World Table 6.1) obtaining similar results and conclusions.2

From column (1) of Table 4 we conclude that the mean growth rate that corresponds tothe five years before a leader dies is statistically different after he is replaced This effect

is even more important when a leader stays more than two years in power As Jonesand Olken (2005), from this column we conclude that leaders matter in average underautocracies but not in democratic regimes 3 Nevertheless, this measure for the real percapita growth rate was subject to many criticisms as pointed out by Johnson et al (2013).These criticisms were based on the divergence of growth rates between PWT releases.One of the reasons was that the real GDP using Laspeyres index in older versions(including PWT 6.1) was calculated as a weighted sum of consumption, investment,public expenditure and exports, where the weights were the share of those components

on the domestic absorption of the benchmark year If those shares change overtime,this technique will lead to biased results The problem due to the change in weights is

2 I excluded leaders Burnham and Jagan Cheddi, both from Guyana for comparison reasons as this country is not available in version 9.0 of the Penn World Table nor in the Maddison Project dataset.

3 The J statistic that I obtained is more significant than the one by Jones and Olken (2005) This can

be due either to the different division of geographic regions or to the four extra leaders included in my analysis.

corrected from version 6.3 on, were they rely on the the total domestic absorption instead

of on the sum of the components (Feenstra et al., 2015) Using this more appropriatedmeasure and keeping the same set of leaders than before, we can therefore see in columns(2) of Table 4 that the main conclusions remain stable

Table 4: The impact of 61 random leaders transitions on growth

However, the 6.3 version still did not solve entirely the problem of inconsistencybetween different versions The biggest dilemma comes from the extrapolation of theparity purchasing power from the reference year to the other ones using relative countries’inflation, as this ignores the difference of countries’ bundle of goods in the computation

of inflation This bias is likely to be higher if the year under consideration is far awayfrom the reference one Yet, according to Bolt et al (2018) “shifts in the bundles ofproducts cannot fully account for these differences, leaving measurement error of somesort as the main (though not very informative) explanation” Thus, when using a singleyear benchmark, the same error is carried through all the data set, and more biased datamake more difficult to detect the significance of a certain effect In more recent releases

of the Penn World Table (beginning in version 8.0) and of the Maddison Project datasetthey introduced a multiple ICP benchmark approach in order to mitigate this issue Themethodology as explained by Feenstra et al (2015) is as follow: “For each country, wekeep track of which benchmarks were used; years in-between benchmarks will have the

prices for final goods interpolated using the corresponding price trends from countriesnational accounts data; and for years before the first or after the last benchmark for eachcountry the prices of final goods are extrapolated using national account data.” Hence,when a new benchmark is available previous growth rates will not change unless thenominal GDP from the national account is corrected.

Using this methodology and keeping the same leaders sample for a proper comparison

we see in columns (3) and (4) of Table 4 that results change: the impact of autocraticleaders is less significant, but more important: national leaders also matter under democ-racies Even more surprising, with PWT 9.0 we conclude that the impact is overallstronger in democracies When redoing the analysis with the Maddison Project data, Iconfirm the existence of an effect under democracies though it is not lasting However,

as presidential terms are likely to be shorter than five years, the treatment timing t + 1and t + 2 will probably contain another transition which would affect the results

Once detecting the bias induced by the growth data used, I explore new leaders’natural death while in power available in more recent data sets.4 Results are presented inTable 5 When using all the leaders included in PWT 6.3 treatment timing t is no longersignificant, which could mean that the effect of a leader takes time to materialize intoeconomic growth rates Unlike previous results, with this dataset we can also detect theeffect of political leaders under democracies Using the Maddison data set (as in PWT9.0) I found that leaders in democracies matter more than in autocracies for economicgrowth Yet, when restricting the Maddison Project to the period 1950 - 2000 in order

to be comparable to the results using PWT 6.3 and with the ones of the previous table,results change drastically and the J statistic is no longer significant in any case Thedifference could be either due to the different countries included or simply to some leaders’idiosyncratic characteristics It is also necessary to recall that not having a significant Jstatistic does not necessarily imply that, in average, leaders do not matter It can be due

to the fact that the countries of leaders we included had a very volatile growth rate and

we therefore did not detect variations on growth pattern or that the qualities of leadersfrom those countries are highly correlated

What we can learn from this analysis is that a leader’s transition can, in fact, have aneffect on growth, but the context in which it may occur is still undefined as changes inthe sample lead to different conclusions Contrary to Jones and Olken (2005), I find that

it is not only a matter of political constraints, as leaders can also matter in democraticregimes The only robust conclusion is that leaders who stayed longer than two years inpower appear to have a stronger impact on growth (either positive or negative)

One of the concerns that may arise with this methodology is that leaders naturaldeaths while in power might not be representative of the overall rulers’ population as

it will be discussed in Section 6 Furthermore, by using those transitions we are not

4 PWT 9.0 does not contain more cases that those presented in Table 4

Table 5: The impact of greater random leaders transitions on growth

distinguishing between the effect of the disturbance on institutions caused by the leader’sdeath and the effect of the leader’s identity (even though this issue is partially mitigated

by excluding the transition year on the Wald Test) In addition, due to the limitednumber of cases it is difficult to analyze how different variables affect the impact thatrulers have on growth and on its direction Indeed, in order to exploit larger cases anotherstrategy is needed

By combining leaders’ data with the national constitutions (Ginsburg et al., 2010),the Cursus Honorum dataset (Baturo, 2016) and the Database of Political Institutions(Beck et al., 2001), we can detect further transitions in which the leader’s identity wasnot likely to be determined by previous growth rates More precisely, I include in thissection those ones in which, by Constitution, the incumbent leader could not run for thesubsequent election Thus, independently of previous growth rates, a new leader has to

take power by a legislative constraint One of the limits of this approach is that electorspreferences may be partially based on the political party So, the candidate that belongs

to the same party of the incumbent leader may be judged by his predecessor’s economicperformance

I also include, when data is available, transitions where the leader enters in office afterwinning elections by a small margin of victory (less than 55% of the votes) consideringthat his victory was random in the sense that his opponent had almost the same chances

to be elected Finally, I add transitions in which the incoming leader was an interim one,

he assumed power by a royal succession, a constitutional one or he was selected by anelite The main assumption is that in those cases the motivations of changing the leaderwere not based on economic conditions I eliminate cases in which, even though one of theprevious conditions held, the leader who exited power did it through an irregular manner(such as coup, revolution or murder) All those cases are represented in the followingselection equation

Θj = 1

(a) if lij−1 died by natural death

(b) if lij−1 could not run for elections due to a constitutional term limit and

lij−1 exited in a regular manner

(c) if lij won the elections by less than 55% of the votes and lij−1 exited in aregular manner

(d) if lij assumed through royal succession, constitutional succession, elected

by an elite or as an interim leader and lij−1 exited in a regular manner

Considering the cases for which growth data is available, 63% of the selected tions correspond to term limits, 28% are the one treated in the previous section and therest correspond to special entry types (cf Table 6)

transi-Table 6: Leaders’ Selected Transitions

Source: Baturo (2016), Ginsburg et al (2010), Goemans et al (2009) and Beck et al (2001)

I first reproduce the same Wald test as before and confirm that leaders matter forgrowth both in autocracies and democracies (cf Table 7) Yet, it is not longer truethat tenure reinforces the leader’s impact Moreover, the volatility of growth under thosetransitions are more prominent in democratic regimes, as found in the last tests of theprevious section This can be due to the fact that in democracies subsequent leaders are

Table 7: Extended exogenous leaders’ transitions and economic growth

to the leader who stayed longer in power Treatment ings t + 1 and t + 2 shift the POST period forward one and two years, respectively Under the null hypothesis, growth is similar before and after a leader leaves office P-values indi- cate the probability that the hypothesis is true The Wald Statistic is the test statistic described in Equation 5 The Chi-squared tests allowed for region-specific heteroskedastic- ity and a region-specific AR(1) process.

tim-more likely to have different characteristics and ideologies than in autocracies where thepool of potential leaders tends to be reduced to a single political party

Having a higher number of cases allows for controlling for further variables Hence,

in the previous approach even if we can run Wald tests for different categories of acertain variable (as done for autocracies or long tenure), it is difficult when it comes

to continuous variables and also when it is necessary to interact several characteristicssimultaneously For this reason, I focus now on the differences of growth rates betweensubsequent leaders in those specific transitions highlighted before To continue with theanalysis of the magnitude of leaders effects, I first regress the absolute value of thisdifference with respect to contextual variables and differences in leaders’ characteristics

as follows:

|∆growth\j−1,j| = α + β∆Xj,j−1+ φ1Yj+ φ2∆Yj−1+ λZj+  if Θj = 1 (9)where ∆growth\j−1,j = is the absolute value of the average growth difference betweenthe growth rates in the first three years of leader j (or less if the leader stayed one or twoyears) and the previous three (or less) years of his predecessor; ∆Xj,j−1is a vector of thedifferences of individual leaders or country’s continuous characteristics; Yj and Yj−1 arevectors of leader or country dummy’s variable corresponding respectively to j’s and j-1’sterm and Zj are control variables such as the initial level of GDP or the country’s level ofdevelopment.5 I also run an alternative regression where I smooth growth rates followingEquations 7 and 8

Results presented in Table 8 show that economies from lower income countries tend

to be more sensitive to political transitions In fact, in those countries the growth ratestend to vary around 10% percent more than in high income countries after a new leadertakes power Potentially, it can be explained by the correlation with the weakness ofpolitical and economic institutions, the regime instability, the level of corruption and

so on In fact, there is also a positive effect when the level of democracy decreases byone point according to the Polity IV score (re-scaled from 0 to 10) This would meanthat when controlling for other variables, the statement that growth variations were morepronounced under democracies does not hold anymore Hence, the relationship betweenthe level of democracy and the effect of the leader is not likely to be linear as the coefficientfor the autocratic dummy is negative

In the selected transitions where the exiting ruler had a possibility to run for reelection,the variation of the growth rate between consecutive leaders tends to be around 3 to 5

% higher Yet, I still do not analyze in which direction This supports the idea that thepossibility of reelection creates incentives to manipulate economic cycles

The relationship between tenure and the impact of a leader was not clear on the vious Wald tests From the first column of Table 8 we infer that under transitions where

pre-a lepre-ader who stpre-ayed more thpre-an two yepre-ars in power lepre-aves the office the fluctupre-ation ofgrowth tends to be lower, contradicting Jones and Olken (2005) conclusions However,when controlling for interaction variables the sign change although it is not longer signif-icant Regarding other individual characteristics, they do not seem to play a role on themagnitude of a leader’s impact In fact, only when a politician leaves office, the growthshift tends to be higher, but this it is only true under the third regression and with an α

of 0.1

In order to analyze the performance of the leaders, I focus on the nominal values

of growth variations between consecutive chiefs of state of the selected transitions (cf

5 The transitional year is imputed to the ruler who stayed more months in power.