DiVA – Digitala Vetenskapliga Arkivet http://umu.diva-portal.org
________________________________________________________________________________________
This is an author produced version of a paper published in Umeå economic studies
Citation for the published paper:
Lilit Hakobyan
Income inequality, competitiveness of political systems and the distance to the efficient frontier of economic growth
Umeå economic studies, 2014, No. 872
http://www.usbe.umu.se/enheter/econ/ues/ues872/
Access to the published version may require subscription. Published with permission from:
Umeå University
Income Inequality, Competitiveness of Political Systems and the Distance to the Efficient Frontier of Economic Growth
Lilit Hakobyan1,2
Abstract: This paper investigates whether and under which conditions democracy renders economic performance more efficient. Efficiency, measured by the ratio of (mean)/ (standard deviation) of output growth, becomes an important indicator of the relative goodness of economic performance when countries face a trade-off between development scenarios with high-mean and low-volatility of output growth. This seems to be a case when economies approach the efficient frontier. However, when countries are far away from the frontier economic efficiency may be improved by simultaneously increasing the mean and decreasing the volatility of growth. This study differs from others on the topic in three basic ways: (i) asymmetric (G)ARCH models are employed to simultaneously estimate the mean and volatility of output growth conditional on the factors of interest; (ii) variations in within-country effects of democratisation on the mean, variance and efficiency of economic growth conditional on cross- country variations of income inequality are analysed; (iii) the asymmetry of deviations from the mean is investigated. The results suggest (do not suggest) that in countries with no (with) military dictatorship history democratisation moves economies towards the efficient frontier. The positive effect of democratisation on the efficiency of economic performance seems to be systematically stronger in countries with lower (higher) income inequality in the countries with (without) consolidated civil governments.
Key words: Mean and volatility of output growth, efficient frontier, political system competitiveness, income inequality, weak institutions, asymmetric GARCH model.
JEL classification: E02, E32, O43, P16
1 Department of Economics, Umeå School of Business and Economics, Umeå University, SE-901 87 Umeå, Sweden (email:
lilit.hakob@mail.ru)
2 The author would like to thank Tomas Sjogren and Jorgan Hellstom for numerous helpful comments.
Introduction
There is a large body of economic literature on the casual effect of democratic qualities of political institutions on the economic performance. This literature investigates the effect of political institutions on both the mean and volatility of output growth. Two main conclusions are as follows.
First, the effect that the quality of political institutions has on the mean of output growth seems to depend on the framework of the research. In the cross-country framework the results seem to be somewhat contradictive.3 In contrast in the framework of within-country analyses the results seem to suggest with less ambiguity that accelerations of economic growth are significantly correlated with political transitions.4 Second, the research seems to clearly indicate that democratic institutions have greater mitigating effect on the volatility of output growth.5 To my best knowledge, research on the effect of the political institutions on the volatility of economic growth is carried out only within the framework of cross-country analyses.
Nonetheless, while in resent years scientific research has made a remarkable attempt to integrate economic growth and business cycle models, in political economics most of the papers investigate mean and volatility of output growth separately. Mobarak (2005) seems to be the single paper, which investigates the effect of political institutions on the mean and volatility of output growth within a system of simultaneous equations. Moreover, the paper argues that there is a negative correlation between the mean and variance of output growth.
If the correlation between mean and volatility of output growth is unambiguously negative, then one may unequivocally conclude that democratic institutions support better economic performance, referring either to the negative correlation between democracy and volatility of output growth, or to the positive correlation between democracy and the mean of output growth. However, the overall picture not seems to be that clear. Both theoretic and empirical research in the field predict somewhat contradicting correlation between the first two moments of economic growth. 6,7 Quinn and Woolley (2001) make an attempt to reconcile recorded contradictions. The authors argue that a
3 For example Barro (1994, 1996), Przeworski et al. (2000), Przeworski (2006) and Mulligan and Sala-i-Martin (2003) did not find any robust effect of democracy on economic growth. However, Acemoglu and Robinson (2004) and Rodrik et al. (2004), using an IV approach concluded that the “historical” quality of domestic institutions play the most important role in the variations in level of income across the countries.
4 Roll and Talbott (2003), Giavazzi and Tabellini (2005), Rodrik and Wacziarg (2005), De Haan and Jong-A-Pin (2007)
5 Quinn and Wooley (2001), Tavares and Wacziarg (2001), Acemoglu and Robinson (2004), de Haan and Klomp (2009)
6 Ramey and Ramey (1995) in their seminal paper overview the channels of positive and negative correlation between the first two moments of output growth. The output growth and volatility may be potentially positively correlated, when the countries as Black (1997) argues have a choice between high-variance, high-expected-return technologies and low-variance, low-expected-return technologies. Precautionary motive for savings (Mirman (1971)) is another argument for a positive correlation between the mean and volatility of output growth. At the same time, under the existence of irreversibilities in investments higher volatility may lead to lower investment, hence lower growth (Bernanke (1983), Pindyck (1991)).
Besides, Quinn and Woolley (2001) argue that “governments worldwide, especially democratic governments employ counter-cyclical fiscal and monetary policies. Various social welfare guarantees have come to be referred to as “automatic stabilizers”. In Schumpeterian and real business cycle models, these policies also dampen the speed of technological innovation, probably reducing long-run growth” (Quinn and Woolley (2001), p.642). In addition, Ramey and Ramey (1991) argue that if firms need to commit to their technology in advance then higher volatility will entail lower meant output.
7 Empirical research in the field finds evidences of both negative (e.g. Ramey and Ramey (1995)) and positive (e.g. Kormendi and Meguire (1985)) correlations.
negative correlation between the mean and volatility of output growth will arise mainly as a result of suboptimal macroeconomic policy. “Black describes how growth and volatility would appear in a relatively efficient set of cases. The competing view describes a world that has major sources of inefficiency. However, in the experience of the real world, both effects are likely present. That is, countries with efficient economies exist side by side with inefficient economies.” [Quinn and Woolley (2001), p. 642]. Hence, on the efficient frontier the countries will inevitably face the choice between the trajectory of economic development with high-growth and high-volatility and one with low-growth and low-volatility. Here and further in this paper, following Quinn and Woolley (2001), the word efficient is used by the meaning it conventionally has in financial literature in the context of CAPM or Markowitz approach.8 This suggests that on the efficient frontier economies have maximum expected growth for varying levels of uncertainty. The efficiency of the economic performance in the current paper is measured by the ratio of observed (mean)/(standard deviation) of output growth, the latter is referred to as an efficiency ratio. Hence, any increase in this ratio will imply that the economy moves towards the efficient frontier. The latter is defined as consisting of the 25 countries with the highest efficiency ratios observed within the period of 1950-2004.
Graph A in Appendix I.2 plots average per capita GDP growth recorded in countries analysed in the current paper versus the standard deviation of output growth observed in the countries during 1950-2004. The right part of the graph suggests clear positive correlation between the mean and standard deviation of output growth near the efficient frontier, which implies that there is a trade-off between the high-mean and low-volatility of output growth closer to the efficient frontier. This trade- off makes simultaneous estimation of the mean and volatility of output growth more instructive in the context of the effect that democratisation has on the economic performance. The question arises whether under democratic governments economies display a performance which is closer to the efficient set. Or, probably most importantly, the question is whether democratisation systematically moves national economies towards the efficient set. In the framework of within-country analysis whenever democratisation demonstrates a mitigating effect on the volatility of output growth to increase economic efficiency it also needs not to affect the mean of output growth. Otherwise one may conclude that democratisation merely increases risk aversion of policymakers.
The current paper simultaneously analyses mean and volatility of output growth in 82 countries during the period of 1950-2004 conditional on political and economic factors. It makes an attempt to investigate the role of democratic institutions on the overall efficiency of economic performance. More specifically it investigates whether and under which conditions democratisation moves economies towards the efficient frontier. Quinn and Woolley (2001) argue that countries with democratic institutions are more risk averse, while the dictatorships are prone to chose scenarios with higher risk. Appendix I.2 seems to support this argument too. However, while the countries with
8 Not surprisingly Black’s theory of general equilibrium is often viewed as an macroeconomic model, which is derived from the basics of CAPM proposed by Sharpe (1964)
strong democratic institutions (i.e., those with highest index of democracy and no political transitions recorded within the period) are mainly concentrated in the bottom of the graph, the upper-right section of the graph includes countries with large variety of democratic qualities of political institutions.
Two patterns of inefficient economic policy, widely recognized in political economy to arise under a democracy with low quality of institutions are clientilism and populism. The first of these phenomena refers to a situation in which goods and services are exchanged for political support.This may include situations, where public employment or allocation of scarce financial recourses (under financial repressions) are used as an income and rent redistribution tool in the politically competitive environment. Clientilism is widely recognised to lead to the systematic underprovision of public goods.9 In contrast populism, which is defined as “the implementation of policies receiving support from a significant fraction of the population, but ultimately hurting the economic interests of this majority”10[Acemoglu et al. (2010)], tends to make public sector of considerably larger size and potentially may lead to long-run macroeconomic destabilizations (e.g., chronically high budget deficits and uncontrollable inflation).
One variable, which may strongly enter into correlation between democratisation and economic performance, is the distributional pattern of income. On the one hand, Robinson and Verdier (2004)11 and Acemoglu et al. (2010) argue that predisposition of countries to phenomena like clientilism or populism may vary along with variations of income distribution. This implies that in those countries, where income distribution makes inefficient economic polices more probable, democratisation may actually push economic performance further from the efficient frontier.
However, on the other hand, intuitively in countries with strong democratic institutions income inequality may be closely correlated with the risk aversion pattern. Hence closer to the efficient frontier, where the trade-off between high mean and low volatility of output growth becomes relevant, those countries where the risk aversion of the population is systematically higher will systematically chose the development trajectories with relatively low mean and volatility. These variations in choices will become more noticeable under more democratic institutions, since according Quinn and Woolley (2001) in democratic (autocratic) counties the governments chose development trajectories which reflect risk aversion pattern of the population (dictator).
9 ”Political scientists make a distinction between policy which is ‘clientelistic’ and that which is ‘programmatic’ and these are conceived of as two polar political strategies that parties or groups contesting power might adopt. On the one hand, political parties can compete for support by offering different types of public goods which affect the entire population. These policies might concern ideological issues, such as human rights, or they may be more economic, such as law and order, trade and macroeconomic policy, or regulatory regimes. On the other hand, instead of focusing on such collective or public goods, parties can concentrate on offering particularistic benefits or private goods to groups of supporters.” [Robinson (2005, p. 9)]
10 “the policy that while being to the left of the political bliss point of the median voter still receives support from that median voter”
[Acemoglu, Egorov and Sonin (2010), p. 1]
11 Alesina , Baqir and Esterly (1998) demonstrated empirically for the case of US cities that public employment is used more heavily as an income redistribution device in the cities with higher income inequality. At the same time, Rodrik (1997) argues that “the (partial) correlation between government employment and exposure to external risk is robust against the alternative hypothesis that government employment has been driven by consideratios of rent-seeking and rent distribution.”
Botswana and Singapore, two countries, which during the second half of the last century were on the efficient frontier of economic development (as it is defined in the current paper), noticeably differ in terms of their levels of initial income inequality and the democratisation of their political systems. Moreover, Argentina and Brazil, which like many Latin American countries, experienced military dictatorships in the 1970 and 1980s, share many common characteristics but differ significantly in terms of income distribution and the size and power of the middle class. In 1964 and 1976, the recorded Gini coefficients of Brazil and Argentina were .5 and .35, respectively. The military regimes established in these countries varied significantly not only in duration but also in economic growth: under the military dictatorship, Brazil grew at an average of 3.2% annually (with 2.32% average annual growth for the period of 1950–2004),12 while the military dictatorship in Argentina during 1976–1983 incurred the cost of an average of -.9% annual “growth”. Does this imply that the effect which competitiveness of the political system has on the economic performance systematically varies along with the variations in income inequality pattern among the countries?
Indeed, in the sample of the 25 countries with the highest efficiency ratio, i. e the countries, which actually define the efficient frontier during the second half of the last century, the Gini coefficient explains up to 64% per cent of variations in the POLITY index (which indicates the competiveness of the political system in countries). At the same time, in the sample of 82 countries included in this paper, the same second order polynomial of the Gini coefficient explains only 16% of political system variation in the OLS regression with a suppressed constant term.13
The current paper investigates the effect of the introduction of a more competitive political system on economic performance measured as the mean and volatility of output growth, as well as its efficiency ratio.
There are three basic ways in which this paper differs from others on the subject. First, asymmetric (Generalised) Autoregressive Conditional Heteroscedasticity ((G)ARCH) models are implemented in an attempt to simultaneously estimate the mean and volatility of output growth conditional on the factors of interest. Empirical research in the field is mainly based on cross- sectional analyses either with observed variance of output volatility included in the mean growth equation, or a system of simultaneous equations including observed average growth and its variance as dependent variables (Mobarak (2005)). GARCH specification of the models, on the other hand, allows simultaneously to analyse the mean and volatility of output growth conditional on the external factors of interest and to exploit information contained in panel data. Analyzing panel data takes into account within-country effect of the democratisation on both mean and volatility of output growth.
For the mean of the growth this may be instructive since the previous research suggests that the
12 However standard deviation of per capita GDP growth in Brazil was 2.7% under the civil political regime and 4.2 % under the military government
13 In general, the following trend is noticeable: while the standard deviation of the POLITY variable is [4.45] in the sample of the 11 countries with top economic performance, [5.24] in the 25 countries and [6.03] in the 82 countries (the total considered in this paper), the explanatory power of the Gini coefficient in a simple OLS model with POLITY as the dependent variable and suppressed constant term is 98% in the sample of the top 11 countries, 63% in the 25 countries and 16% in the whole sample.
results may considerably differ in the context of within or between country analyses. As for the volatility of output growth conditional on the political institutions, to my best knowledge this is a first attempt to evaluate volatility of output growth conditional on the change in political institutions.
Moreover, additional to panel data, EARCH models are implemented for those countries, where political transitions are recorded within the period, and the coefficients of within-country effect of democratisation on the mean, volatility and efficiency of output growth are further regressed on cross- sectional variables.
Second, the effects that democratisation has on the mean, expected variance and efficiency of economic growth is regressed on the distributional pattern of domestic income. This, hopefully, sheds a light on the question whether democratic political systems are economically efficient in countries with one distributional pattern of the domestic income, while dictatorships are more economically efficient in the others. This question seems to have gained an increasing interest of scientists within resent years.14 Further, following Mobarak (2005) to check the robustness of the correlation between income distribution and the effect that competitiveness of political system has on the mean and volatility of output growth a number of macroeconomic factors (such as the composition of domestic production, financial market development etc.) is additionally controlled for.
In addition, the analyses separate countries with and without history of military dictatorship (MD). Acemoglu and Robinson (2006(a)) and Acemoglu et al. (2010) argued that under significantly high income inequality, the nascent (renascent) democracy in countries will remain unconsolidated—
implying a higher probability of coups d’état, on the one hand, and less redistributive economic policy under democratic regimes, on the other hand.15 At the same time, Quinn and Woolley (2001) argue that while the longevity in office may be an important factor encouraging autocrats to risk seeking economic policy, non-constitutional entry to executive office increases the risk of losing power.16This implies that analysing the countries with and without history of military dictatorship separately may be justified for the purposes of the current paper.
Third, the asymmetry of deviations from expected growth rates is analysed conditional on political variables. Here two main questions are focused: i) whether symmetric reaction to good and bad economic news is an attribute of more democratic (authoritarian) regimes; ii) which of the
14 For example, Djankov et al. (2004) argued that under increasing income inequality, the efficiency of the regulating institutions with greater dictatorship and lower disorder will increase, which will tend to make the political institutions having more elements of dictatorship economically more efficient. In contrast, Acemoglu et al (2007) argue that in the environment of weak political institutions high economic inequality may ensure better property rights.
On the other hand Engerman and Sokoloff (2002, 2005) showed that under significantly higher income inequality, economically inefficient institutions evolved in the New World. Rodrik (1997), in a study of a quality of institutions in a sample of East Asian countries, found a strong negative correlation between the quality of institutions and the Gini coefficient. This may also imply that democracy is economically more efficient countries with lower income inequality, if inefficient institutions are more pervasive for the economic performance in the democratic rather than autocratic framework.
15 At least two of the mentioned theories (Robinson and Verder (2004) and Achemoglu and Robinson (2006(a))) predict a potentially nonlinear relationship between income inequality, democracy and economic outcome.
16 Alternatively, Rodrik (2000) suggests that in the political system where power frequently switches hands and any of the competitive sides have sufficiently high probability to return to the office after being deposed, the higher degree of preferences heterogeneity with respect of public goods provision with higher probability will lead to cooperation, i.e., to less volatility of economic policy. In contrast, in both stabile and instable dictatorships, where (i) the power either does not change hands frequently or (ii) the power shifts occur frequently but the ruler has no chance of coming back once deposed, the higher heterogeneity of preferences will lead to the higher volatility of economic policy.
political regimes is more capable to utilise favourable economic shocks (i.e. under which political regimes economies demonstrate stronger reaction to favourable rather than unfavourable shocks).
This paper is organised as follows. Section Two presents the data and descriptive statistics.
Section Three describes the econometric model. Section Four discusses the results. Finally, section Five concludes the paper.
2. Data and Descriptive Statistics
Output time-series data are taken from “Historical Statistics for the World Economy: 1-2004 AD,” Maddison (2008). The dataset provides uninterrupted time-series of per capita GDP for all of the countries under study for the period 1950–2004. Recorded per capita GDP is measured in 1990 US dollars converted at purchasing power parities.
The efficiency of economic growth is measured by (mean)/ (standard deviation) of per capita GDP growth ratio, labelled as efficiency ratio. The larger the ratio is the closer the country to the efficient frontier of output growth is considered to be. The frontier is aimed to plot the most efficient combinations of mean and volatility of output growth available worldwide in 1950-2004 (i.e.;
maximal additional expected growth for the additional points of volatility). Such a frontier should exist since the correlation between the mean and volatility of growth in countries with most efficient economic performance proved to be negative. By the definition the frontier should take into account only the countries which demonstrate economic performance not dominated by the performance of other countries in the space of mean-standard deviation of output growth. This will include only Norway, Italy, Austria, Ireland, Spain, Japan, Taiwan and South Korea. However, this definition of the frontier has two considerable drawbacks. First, both table I.1 and graph (a) in appendix I.2 imply that Norway and Taiwan look as complete outliers (other two outliers seem to be Spain and Japan).
Second, as it is implied by the list of the eight countries mentioned above, the frontier will have rather restricted cultural and geographic composition. To demonstrate efficient economic performance available worldwide, the frontier should include countries with diversified cultural-geographical background and different states of technological development at the beginning of the period. Hence, certain trade-off arises between the notions of efficient performance and worldwide availability. By another approach the frontier may be approximated by a group of countries with the highest recorded efficiency ratios. Because of the definition of the efficiency ratio, this group of countries in OLS regression of the mean growth rate will demonstrate the largest slope with respect to its standard deviation. In addition, the smaller the group of countries with consecutive efficiency ratios is the higher predictive power OLS regression will gain. Table I.2 in appendix reports R2s and slopes of OLS regressions for the groups of 10 to 40 countries with highest observed efficiency ratios. The table shows that up to the size of 25 countries any 5 countries additionally included into the groups
cause less than 5% lose in R2, while starting from the group of 30 countries inclusion of additional countries lead to R2 shrinking with growing speed. This consideration makes the group of 25 countries with the highest recorded efficiency ratios become the best candidate for approximation of the efficient frontier.17 Graph (b) in appendix I.2, which replicates graph (a) with growth ranks instead of the names of the countries, indicate that both Botswana and China, with growth ranks of 24 and 25 respectively, should be considered to be on the frontier, as long as South Korea is treated as an outlier. Hence the frontier is defined to consist of the group of 25 countries with recorded highest efficiency ratios18, 19 though the group includes countries with economic performance dominated by the other countries in the group.20
The group of countries with efficiency ranks of 26 to 30 is also of special interest for two considerations. On the one hand, the group, which includes Colombia, Malaysia, Israel, India, and Egypt,whenever added to the 25 countries of the frontier will make the frontier considerably larger in terms of cultural and geographical diversity. On the other hand, these countries seem to demonstrate sound economic performance during the second half of the last century. 21 In addition, the trade-off between scenarios with high-mean and low-volatility of output growth is still clearly present within this group of countries (tableI.2 indicates that with addition of these countries only .04 paints is lost in the slope OLS regression). Hence the larger concept of the frontier includes these five countries, disregarding the fact that all of them demonstrated economic performance strictly dominated by number of countries with ranks of 1 to25. Graphs (c) and (d) in Appendix I.2 plot mean and standard deviation of the 25 and 30 countries of the efficient frontier.
Political data are taken from the Polity IV database, which is widely used in empirical research concerning political institutions. The database contains coded authority characteristics states for comparative and quantitative analysis. It includes constructed annual measures of both institutionalised democracy (DEMOC) and autocracy (AUTOC).22 The measures are composite indexes derived from the coded values of the following component variables of authority characteristics: regulation of chief executive recruitment, competitiveness of executive recruitment,
17 One may expect the relationship between the mean and standard deviation of output growth to be convex (with ever growing additional volatility required for additional expected growth, when the latter becomes larger) rather than linear. Moreover, graphs (a) and (b) in Appendix I.2 seem to support these expectations. Graph (b), nonetheless, indicates that linear approximation may work well as long as Botswana or China are considered to lie on the frontier, and South Korea is regarded as an outlier.
18 This includes fourteen Western Europe countries (growth ranks in parentheses): Norway(1), France (2), Italy(4), Ireland (5), Belgium (6), Austria (7), Spain (8), Sweden (11), UK (12), Greece (13), Netherlands (14), Portugal (15), Denmark (18), Finland (20); three out of the four Western Offshoots: Australia (10), Canada (21), USA (22); four Pacific Asia “Tigers”: Taiwan (3),Thailand (16), South Korea (17), Singapore (19) and Japan (9); another fast growing Pacific Asia countries: China (25); and a single country from: Africa- Botswana (24)
19 In the list of countries which are left out of the sample in this study Germany seems to be the only one which has high probability to be included in the list of countries with most efficient economic performance for the period under consideration.
20 Taiwan dominates Thailand, Singapore , China and Botswana; Japan—Thailand, Singapore and China; Spain and Ireland both –Finland and Portugal; Norway demonstrates economic performance superior to the performance of the bunch of countries, included into the frontier due to observed low volatility of output growth. So that in its narrowest definition the frontier would include only Norway, Italy, Austria, Ireland, Spain, Japan, Taiwan and South Korea.
21 The main surprise here seems to be Colombia. The country enters the list mainly due to an extremely low volatility of output growth recorded in the period.
22 The logic of this "institutionalised autocracy" scale is similar to that of the “institutionalised democracy” scale. However, the two scales do not share any categories in common. Many polities have mixed authority traits and, thus, can have middling scores on both the Autocracy and Democracy scales. For more detailed information, see the POLITY IV manual.
openness of executive recruitment, executive constraints (decision rules), competitiveness of political participation, and regulation of participation. “A mature and internally coherent democracy might be operationally defined as one in which (a) political participation is fully competitive, (b) executive recruitment is elective, and (c) constraints on the chief executive are substantial’’ [Polity IV Project:
Dataset Users’ Manual, p. 16]. The POLITY index is calculated by subtracting the AUTOC value from the DEMOC value. The procedure provides a single regime score that ranges from “+10” (full democracy) to “-10” (full autocracy).
The current paper, distinguishes between weak political institutions and the competitiveness of executive and legislative bodies in elections. The paper makes an attempt to look into how the choice of inefficient policies under increasing political competition and weak institutions systematically varies among the countries, depending on country-specific characteristics such as income inequality.
Such a systematic variation may imply that a more competitive political structure may be economically efficient (i.e. put the economy close to the frontier) under one income distribution and relatively inefficient under another distribution. The POLITY index is used as a proxy of political system competitiveness. Two variables are used as proxies of the weakness of political institutions:
the Executive Constraint Index and the Number of Political Transitions.23 Both of these variables are taken from the Polity IV database.
Executive Constraint index is one of the six components of POLITY index, which is often used in empirical research as an indicator of the quality of political institutions. “Operatioally, this variable refers to the extent of institutionalized constraints on the decision making powers of chief executives, whether individuals or collectivities. Such limitations may be imposed by any
"accountability groups." In Western democracies these are usually legislatures. Other kinds of accountability groups are the ruling party in a one-party state; councils of nobles or powerful advisors in monarchies; the military in coup-prone polities; and in many states a strong, independent judiciary.
The concern is therefore with the checks and balances between the various parts of the decision- making process.” [Polity IV Project: Dataset Users’ Manual, p. 24] A seven-category scale is used for the index, where the category 1 refers to the situation with “unlimited authority” and category 7 to the well-balanced executive power and strong control.
The Number of Political Transitions is calculated as the total number of the times, when a unit point or larger scale changes are recorded in the POLITY index, i.e. the total number of times, when political transitions are recorded, in the period 1950–2004. It is used as a proxy of the weakness of political institutions for the following reasons. First, political institutions mainly matter for the reason that it is easier for some social groups to solve the collective action problem in some periods than in other periods. Thus, there is a need to use a temporary strength to redesign political institutions to
23 One major problem associated with the use of Executive Constraint Index is the high correlation between this variable and POLITY index (the latter is used as a proxy for political competitiveness) the coefficient of correlation between the two variables is .76.
ensure future benefits.24 This implies that the stronger and better organised the social parties in a country are, and the more evenly over the time the power distributed is, the less relevant the design of political institutions becomes. The vast majority of OECD countries had no political transition in the second half of the last century. Second, a large number of political transitions along with the experience of military dictatorship(s) indicates unconsolidated democracy, where both political elites and the general population only become powerful episodically and need to redesign political institutions to ensure the future enforcement of their preferences.
The list of military dictatorships is taken from the Integrated Network of Social Conflict Research Data Page: Coup d’état 1946-2009, which is the part of the Polity IV project.
All countries that had cases of “foreign interruption” or “interregnum or anarchy” within the period 1950–2004, as well as former socialist countries, are excluded from the analysis. Former socialistic countries are excluded because during a significant part of the period under consideration the volatility of output growth in these countries was conditioned by the structural changes that took place in national economic systems (in a number of cases regardless of political transformations and out of the control of the central government). Hence, within the context of these countries efficient macroeconomic policy more probably will imply efficient transformation and crisis management rather than economic policy orientated towards the efficient frontier and the optimal choice between high-mean high-volatility and low-mean low-volatility paths. In addition (with exception of Albania, Bulgaria, Hungary, Poland and Romania) for this group of the countries the data are available only starting from 1990, which means that the time-series are too short for ARCH model specification based analyses. In countries with “interregnum or anarchy”, on the other hand, it seems plausible to expect that the central government may fail not only to conduct efficient macroeconomic policy but also to efficiently conduct macroeconomic policy for reasons other than the competitiveness of the political system and distribution of national income.
Income distribution data are taken from the World Income Inequality Database. The database contains composite income distributional data from various sources. For all of the sample countries, gross income data covering all regions and ages with no household adjustments are used. Income inequality is treated as a predetermined variable and is taken by the value recorded at the beginning of the period, i.e. in1950. In those cases where no data were available for a given year, the gross income distribution entry for the nearest available year was used. Such a treatment is supported by the conventional belief that the variable usually has quite a slow dynamics, which makes between- country variations more considerable than within-country ones. This implies the effect that income inequality has on the mean, volatility and efficiency of output growth should be interpreted in a cross country framework only. However, such a treatment helps to overcome shortcomings which the inclusion of income inequality into dynamical panel potentially entails. The first of these
24 Acemoglu et al. (2006) single out Germany, where the working class had the best and most powerful organization in Europe and where democratization took place after France and the UK, probably because the working class was better organized and could enforce its preferences even without formal changes in political institutions.
shortcomings concerns the missing data. The second shortcoming is related to the potential edogeneity problem, which may arise if income distribution is accounted for as a dynamic variable.
Gross income distribution is preferred to the distribution of after-tax income because of the belief that the latter, while serving as a better indicator of redistributive preferences of overall population, at the same time may smooth out the conflict initially existing between the incomes, i.e.
between the redistribution preferences of the different groups of population. Hence, while distribution of gross income is a potential cause of the efficiency (non-efficiency) of the chosen macroeconomic policy the distribution of the net income is most probably the outcome of the chosen policy. However, to check for the robustness of the correlation between the distribution of domestic income and the effect that the competitiveness of the political system has on the economic growth additional macroeconomic factors such as the size of the government, share of agricultural goods, domestic credit to private sector are controlled for. The following variables: Share of agricultural goods, Investment share of GDP, Share of government expenditures, Share of export plus share of import in GDP, Manufactured goods share in export, Fuel share in export, Central government debt, External debt stock, Inflation index of consumption goods, Broad money ratio to total reserves, Domestic credit to private sector and Secondary school enrolment are taken from the World Bank Database:
World Development Indicators (WDI) & Global Development Finance (GDF).
Table 1: Descriptive statistics
Variable Mean Std. Dev. Min Max
Per Capita GDP
Log of Per Capita GDP at the Beginning of the Period 7.4768 .9813 5.7149 10.3217 Average Growth Rate of per capita GDP in 1950-2004 2.0651 1.4796 -1.8451 5.7564 Standard Deviation of GDP Growth in 1950-2004 4.2883 2.1804 1.6359 14.3849 Mean /Standard Deviation Index of Per Capita GDP Growth .5992 .4624 -.1980 1.8525
Income Distribution
Gini Coefficient in 1950 .4619 .0920 .2910 .6890 Middle Class Income/Rich Income in 1950 .4456 .1332 .1765 .7046
Political Institutions
Average POLITY index in the Period 1.5127 5.9696 -10 10 Average Executive Constraint in the Period 4.0621 2.0148 1 7 Number of Political Transitions in the Period 3.1708 2.8838 0 11
Sample Countries
Number of Countries 82
African Countries 23
Arabic Countries 13
Latin American Countries 16
OECD Countries 18
Pacific Asian Countries 12
OPEC Countries 13
Countries with Colony Status Before the End of WWII 32
Countries with no Political Transitions in 1950-2004 14 Countries that Experienced Military dictatorship (MD) in 1950-2004
34
Moreover, Acemoglu and Robinson (2004) argue that in certain cases, mean-preserving income inequality measures such as the Gini coefficient may not be good at explaining the de facto distribution of political and economic preferences, a proxy of (Middle Class Income) /(Upper Class Income index (MCUCI) was used as an alternative. The MCUCI index was calculated as (D4+D5+D6+D7)/(D8+D9+D10), where D4 to D10 are the fourth to tenth deciles of income distribution. Hence the index accounts for the relative income of the middle class and upper class, and the income inequality increases as the index decreases.
Data on economic openness and liberalisation are taken from Wacziarg and Welch (2008) and Sachs and Warner (1995).
Table 1 provides descriptive statistics on the key variables used in this paper. The data include 82 countries (a list of the countries is reported in Appendix I) and 54 years between 1950 and 2004.25
Appendix II graphically demonstrates the relationship between the mean, standard deviation, and (mean)/(standard deviation) ratio of per capita GDP, one the one hand, and income distribution and politico- institutional factors, on the other hand. Table 2 summarises the OLS regressions run with single independent variables. The graphs and regressions suggest that income distribution has more power in explaining variations in the mean of per capita output growth, while institutional factors are better at explaining the cross-country variations in standard deviations. Nevertheless, all of the variables seem to have strong explanatory power concerning variations in the (mean)/(standard deviation) ratio of output growth across the countries, which roughly indicates the distance to the efficient frontier.
Table 2: Correlation between economic growth and country-specific factors Dependent
Variable Gini Coefficient
at the Beginning of the Period
Middle Class Income/Rich Income at the Beginning of
Period
Average POLITY index within Period
Average Executive Constraint within Period
Number of Political Transitions within Period Average Growth Rate in
1950–2004 Adjusted R2
-6.04***
.1479
5.46***
.2614
.0612**
.0627
.224***
.0824
-.456**
.0446 Standard Deviation of Per
Capita GDP Growth Adjusted R2
3.83**
.0449
-3.97***
.1138
-.203***
.3697
-.646***
.3488
.558*
.0272 Mean/Standard Deviation
Ratio of Per Capita GDP
-2.36*** 2.14*** .045*** .151*** -.296***
25 In cross-sectional analyses POLITY and Executive Constraint indices are taken by average value they took within the period, while income distribution variables are taken by the value they have at the beginning of the period (i.e., in 1950). This is done for two reasons.
First, because annual entries for the former variables are available, while for a large number of the countries only couple of entries are available for the whole period of 1950-2004. Second, income distribution proves to be much more slowly moving factor than the POLITY index is (especially in countries with MD experience).
Growth Adjusted R2
.2144 .3685 .4033 .4280 .2341
The OLS model is run with constant and a single independent variable; *, **, *** indicate that the coefficient is statistically significant at the 10%, 5%, and 1% levels, respectively
3. Econometric specification and estimation
This paper is focused on the following three question: i) the effect that the competitiveness of political system has on the mean, expected variance and efficiency of output growth; ii) systematic variations of the within-country effect that transition to more competitive political system has on the economic performance conditional on cross-country variations of the income distribution; iii) the effect that the quality of political institutions and income distribution has on the asymmetry of deviations from the mean of output growth. Investigation of the first question is mainly based on panel data analyses. Investigation of the effect that the quality of political institutions has on the efficiency of economic performance and the second question, in contrast, mainly focus on cross- country variations of within-country effect that democratisation has on the economic performance.
Finally, the third question focuses on the cross-sectional differences of dynamical pattern of per capita GDP growth for individual countries depending on the cross-sectional characteristics of the countries.
In an attempt to simultaneously estimate mean and volatility of output growth time series Asymmetric Autoregressive Conditional Heteroscedasticity models (GJR-(G)ARCH and E(G)ARCH) were implemented to test the effect of institutional factors and other factors of interest on the conditional mean and volatility of output growth time series.
Although ARCH models are typically used to forecast uncertainty in time series of stock returns, foreign exchange and interest rates, Engle (1982) first applied his resulting ARCH model to parameterise conditional heteroscedasticity in a wage-price equation in the United Kingdom.
According to Engle (2004), the original idea was to find a model that could assess the validity of Friedman’s (1977) conjecture that the unpredictability of inflation was the primary cause of business cycles.
(G)ARCH models allow for simultaneous estimation of mean and volatility of output growth time series conditional on institutional factors. At the same time, empiric research in political economy supports the idea that the quality of political institutions may simultaneously affect the first and second central moments of GDP growth time series. Moreover, Ramey and Ramey (1994) and Mubarak (2005) found evidences that high volatility deters the mean growth of countries’ output time series. However, the question of possible correlation between institutional factors and the third central moment of GDP growth time series, i. e. symmetry of the distribution around the mean, remains open.
Appendix I of this paper shows the data on the skewness of GDP growth series for the sample
countries. In majority of the cases lower tail of the distribution seems to be longer than the upper one.
This suggests that for the majority of the countries downward deviation from the expected growth rate is larger than upward deviation.26 Employment of asymmetric (G)ARCH models allows for the further analyses of the asymmetric deviations and lengths of the upper and lower tails of growth series conditional on the institutional factors.
In its simplest framework the GARCH (1,1) model may be presented as follows:
yt t htzt (1) ht t zt21 ht1 (2)
where {zt} is a (iid) random variable with a mean of zero and unit variance; yt = ln (yt) - ln (yt-1) with ln(yt) being the log of per capita GDP and yt the first difference of the log of per capita GDP; t is the mean of growth rates conditional on a set of exogenous factors x j,t and may be written as follows:
jt
n
j j
t b b x ,
1
0
(3)
Equation (2) describes the variance of output growth rates and to include additional explanatory variables may be rewritten as:
n
j
j t j t
t
t z h x
h
1 , 1
2 1
0
(4) Where
n
j
j t j
t x
1 ,
0
.
The conditional variance is determined by a constant, set of exogenous volatility regressors xj,, prior shock z2t-1 as an ARCH term and past variance ht-1 as a GARCH term. The term λjxj captures the effect of the exogenous factors on the volatility of output growth time series. If an exogenous factor xj has a coefficient λj, which is significantly different from zero, then the variation of factorxj
will have a permanent effect on the volatility of growth time series. The ARCH term, on the other hand, captures the tendency of volatility clustering, i.e., small/large deviations from the conditional mean result in small/large changes. This implies as α increases, the response of the economic system to large shocks becomes stronger. In the model, which describes the dynamics of output growth (like in the models describing the dynamics of financial indices) this term may indicate the response of the economic agents to the information about realisation of prior expectations. The GARCH term captures persistence in volatility. It may be interpreted as a process of adaptation to permanent and temporary changes in variance. If conditional volatility is defined only by the ARCH component it vanishes rapidly unless the term is included by too many lags. With the GARCH component, this effect vanishes more slowly, though still exponentially. A large β coefficient indicates that the system
26 However, under the condition of unimodal distribution of growth time series this will also imply that the upward deviations are more frequent than the downward ones and that the median lies above the mean
has a memory that is longer than one period. When β is larger than α, then the volatility is more sensitive to its own history than to new shocks. The long-run average variance then will be defined as
) 1
/(
.
The model can be estimated via maximum likelihood, where the distribution of the ht-1z2t-1 innovations is assumed to be Gaussian.
Equation (4) implies a symmetric response to both positive and negative deviations from expected growth. However Appendix I suggests that distribution of growth data around the mean growth rates is not symmetric in the countries. Besides, Nelson (1991) argues that symmetric GARCH models do not allow for the oscillating dynamics of adjustment process. There are a large number of (G)ARCH based models that are adjusted for response asymmetry. In this paper, two of these models GJR-(G)ARCH and E(G)ARCH, are applied.
The GJR-GARCH model originally proposed by Glosten, Jagannathan and Runkle (1993) substitutes the conditional volatility equation (4) with the following:
n
j j t j t
t t
t I z z h x
h
1 , 1
2 1 1
0 ( 0)
(5)
Where I (zt <0) is an indicator function that takes the value one if the condition holds, i.e., when the prior shock has a negative sign, and zero otherwise. Equation (5) implies that a positive shock will affect the volatility with the coefficient α, and negative shock will have the coefficient (α+γ).
Another asymmetric GARCH model that is applied in this paper is EGARCH. The model introduced by Nelson (1992) describes conditional variance using the following equation:
n
j
j t j t
t t
t z E z h x
z h
1 , 1
1 1
1 0
1)
ln( (6)
Equation (6), which has (γ- α) as the coefficient of a response to a negative shock and (α+γ) as the coefficient of a response to a positive shock, is asymmetric in zt-1. The system responds to negative shocks with higher volatility if α>0 and γ<0.
Since the current paper is interested in the effect of political institutions on the mean and volatility of output growth time series, the central problem becomes the robustness of results (rather than the accuracy of prediction). Hence, different models were employed alternatively to check robustness.
A dynamical panel with unique slopes for factor variables across the sample countries was estimated using GARCH, GJR-ARCH, EARCH, GJR-GARCH specifications, and t is assumed to have a Gaussian distribution.27 Pooled models are specified since some of the factors of the main focus (such as POLITY index) do not vary in number of the sample countries within the period under consideration, which means that under the fixed effect specification the effect of these variables will be captured by the countries’ intercepts. The log of per capita GDP at the beginning of the period
27 Additional analyses are implemented with models where zi is assumed to have t-distribution. The results are not reported since do not considerably differ from those with assumption of Gaussian distribution
(i.e. in 1950), which is included into specified models, is aimed to (partially) capture the effect of omitted country specific factors correlated with both dependent variables and factors of main interest.
The first set of questions concerns the effect that the quality of political institutions has on the mean and volatility of output growth. The main variables of focus here are POLITY index, the Number of Political Transitions and the Executive Constraint index. The former of these variables is used as a proxy for competitiveness of political systems, while the latter two are used as proxies for the weakness of the political institutions. The specified models are estimated as dynamical panel in the pooled sample of 82 countries, and separately in the sample of the countries with and without history of Military Dictatorship (MD). Following the insight of Acemoglu and Robinson (2006(a)) and Acemoglu et al (2010) it seems plausible to expect that the effect, which the competitiveness of political system has on the economic performance of the countries may significantly differ among the countries with and without MD history.
Endogeneity of the right-hand variables is a common and recognized problem in model specifications simultaneously including output growth and institutional quality, which means that the results may be seriously biased if the quality of political institutions is, in turn, affected by the mean and volatility of expected output growth.28To overcome the problem Acemoglu et al (2004) instrumented the quality of political institutions using “mortality rate of European solders during colonization period”. While this factor is recognized to be a good instrument it has two limitations.
First, it is applicable only to the list of countries which were subject to colonization. Second, the instrumented variable(s) may be used only in cross-country analyses. The instrument may be effectively used to predict the quality of political institutions, which persist due to the persistence of inefficient institutions over the long periods of time regardless the changes in de jure political system.
The current paper, on the other hand, is mainly interested in the competitiveness of de jure political systems. The latter variable displays impressive variations in a number of sample countries (especially in the group of countries with MD history) within the period under consideration. This makes instrumentation of the “mean POLITY index” observed in countries within the period meaningless for the purposes of this paper.
To overcome the problem VAR (1,1,1) (vector autoregressive) model is additionally estimated to check the robustness of the conclusions to the potential endogeneity of political institutions to the mean and volatility of output growth. Estimated VAR(1,1,1) models have the following three variables as the dependent variables: POLITY index, five year rolling average of per capita GDP growth and rolling five year standard deviation of per capita GDP growth. The models are alternatively estimated for the pooled sample of all countries and subsamples of countries with and without MD history. Because in VAR specified models all the three variables are simultaneously endogenous, estimation of the models gives an opportunity to account for the effect that the dynamics
28 This arguments seems to be plausible and has its advocates in political economic literature
of economic performance (i.e., mean and volatility of output growth) has on the competitiveness of political system (i.e., POLITY index).
The second set of questions focuses on the cross-country variation of income distributional pattern in the countries and the effect that competitiveness of political systems has on the mean and volatility of economic growth. Here the main variables of focus are the Gini coefficient, (Middle Class Income)/(Upper Class Income) index, labelled as MCUCI, and interaction of POLITY index with these two variables.
Two alternative methods are employed. First, GARCH and GJR-GARCH models are estimated in a panel of countries with four explanatory variables: log of per capita GDP at the beginning of the period, POLITY index, Gini coefficient or MCUCI and interaction of (Gini Coefficient)*(POLITY index) or (MCUCI)*(POLITY index).
Since the models control for the income inequality indices at the beginning of the period there is no simultaneity problem rising from the specifications. However, still there may be an omitted variable problem present in the results if all the three factors (i.e.; income inequality, competitiveness of the political system and output growth volatility) in the countries are correlated with the (omitted) fourth factor, such as risk aversion. If higher income inequality in countries reflects lower risk aversion, one may expect that under democratic institutions more risky scenarios of the economic development will be chosen by the median voter. 29There are (at least) to line of the arguments supp
Hence if the risk aversion pattern is strongly correlated with income inequality one may expect this correlation to become more noticeable: i) under the higher competitiveness of political institutions; ii) closer to the efficient frontier of output growth. The former expectation arises due to the argument of Quinn and Wolley (2001), which states that democratic governments chose development scenarios which mainly reflect the risk preferences of the population, while in authoritarian countries development strategies mainly reflect the risk aversion pattern of the dictators.
The second expectation will arise only when there is a trade-off between the high mean and low volatility of output growth close to the efficient frontier. To check this hypothesis further the sample is split into two following subsamples: i) 30 countries with the highest observed efficiency ratio (i.e., the countries which define or are very close to the efficient frontier); ii) the rest of the countries.30
29 There are (at least) two lines of arguments, which support the intuition that in democratic countries higher income inequality may reflect lower risk aversion of the population. First, if in countries with strong democratic institutions economic relations are close to free competition (this argument is in line with Acemoglu and Robinson (2008)), then in countries with lower risk aversion the population will choose projects with higher risk than in countries with higher risk aversion. One may expect that by the passage of time this higher volatility of realised outcomes will lead to faster diverging paths of accumulated capital among the population. This, in tern, will be translated into higher income inequality. Second, democratic countries, where the population has higher risk aversion most probably will choose heavier redistributive policy. This in tern will lead to lower income inequality.
30 While the efficient frontier is defined by 25 countries, the subsample includes 30 countries for the following considerations. First, 25 countries on the frontier mainly include European countries or Western Offshoots, which may imply limited variations in both political culture and income distribution pattern. Second, the 5 countries additionally included into subsample clearly demonstrate that the trade-off between scenarios with high-mean and low-volatility of output growth is still relevant. Nonetheless, additional analyses are implemented for the subsample of 25 countries. The results are close to those based on the subsample of 30 countries (and reported in this paper) but somewhat less impressive.
