Political Instability and Economic Growth

This paper investigates the relationship between political instability and per capita GDP growth in a sample of 113 countries for the period 1950-1982. We define ?political instability? as the propensity of a government collapse, and we estimate a model in which political instability and economic growth are jointly determined. The main result of this paper is that in countries and time periods with a high propensity of government collapse, growth is significantly lower than otherwise. This effect remains strong when we restrict our definition of ?government change? to cases of substantial changes of the government.


Introduction
Economic growth and political stability are deeply interconnected. On the one hand, the uncertainty associated with an unstable political environment may reduce investment and the speed of economic development.
On the other hand, poor economic performance may lead to government collapse and political unrest.
This paper studies the joint determination of the propensity of government changes (our measure of "political instability") and economic growth in a sample of 113 countries for the period 1950-

1982.
The primary result of this paper is that in countries and time periods with a high propensity of government collapses, growth is significantly lower than otherwise. This effect is strong for both of the two types of government changes considered: all government turnovers including those that do not involve a significant change in ideological direction or an "irregular" transfer of power or alternatively those government turnovers that involve only these two types of changes.
Our other results are: 1) Contemporaneous low economic growth is not found to increase the contemporaneous propensity of government changes. 2) We do not find any evidence that economic growth is significantly different when authoritarian regimes are compared to democracies. 3) We find that political instability tends to be persistent, in that the occurrence of frequent government collapses increases the probability of additional collapses. This is not the first paper which studies the relationship between economic outcomes and political instability in a large sample 1 of countries. * Notably, Barro (1991) finds that measures of political unrest, such as number of assassinations and the occurrence of violent revolutions and military coups significantly affect the average growth level in cross section regressions on a large sample of countries. In addition, Kormendi and McGuire (1985) and Barro (1989) find that a measure of the extent of political rights is positively correlated with growth.
Other studies which have adopted a notion of political instability similar to ours have found effects of instability on inflation (Cukierman, Edwards and Tabellini (1992)), and on external borrowing (&ler and Tabellini (199 1)). In these papers, a measure of political instability is added as a regressor in cross section or panel regressions; however, this methodology does not take into account the joint endogeneity between the economy and the polity. Poole (1990, 1991a) have addressed this problem of joint endogeneity by estimating a system of equations in which the dependent variables are GNP growth and coups d'etat.2 Their results are different from ours, in particular in that they do not find evidence of reduced growth as a consecuence of increased political instability. Our study differs from theirs in several important respects.
First, in model specification. In our growth equation we control for several economic determinants of growth, as identified by the recent empirical literature on economic growth, as well as some indicators of political unrest in the government change equation. In addition, we do not primarily focus on "coups d'etat" but on a broader definition of government changes, which includes not only coups but also 2 constitutional transfers of power: political uncertainty is not confined to the occurrence of military coups. We present some results based on our specification focusing only on coup detat so as to suggest the sources of differences in the results of these two sets of works.
This paper is organized as follows. Section 2 discusses the basic questions which we explore in our empirical analysis. Section 3 presents the econometric methodology used both for the by now familiar in the cross-sectional methodology and for the joint estimation of the growth and government instability equations using panel data.
Section 4 describes our data set, highlights some basic statistics and describes the specification used in the estimation of the cross-section and panel models. In Section 5 we present our cross-section regressions of growth. The main results of the panel estimation of our simultaneous equation system are discussed in Section 6. Section 7 presents alternative specifications so as to investigate the robustness of our results. The last section is a discussion of future avenues of research.
2. Does political instability affect economic growth?
The first step toward answering this question is a definition of what it is meant by "political instability". In this paper, "political instability" is defined as the propensity of a change in the executive, either by "constitutional" or "unconstitutional" means. Thus, we study whether a high propensity of an executive collapse leads to a reduction of growth.
3 One strong theoretical argument underlying this relationship is based upon the effects of uncertainty on productive economic decisions, such investment, production or labor supply? A high propensity of a change of government is associated with uncertainty about the new policies of a potential new government; risk-averse economic agents may hesitate to take economic initiatives or may "exit" the economy, by investing abroad. Conversely, foreign investors prefer a stable political environment, with less policy uncertainty and less uncertainty about property rights4 Alesina and Tabellini (1990), Tabellini and Alesina (1990), Cukierman, Edwards and Tabellini (1992), 6zler and  present several models in which a government is uncertain about its survival, and as a result engages in suboptimal policies in order to "worsen" the state of the world inherited by its successor.' All these models have in common the idea that political instability lead to economic inefficiencies. 6 The most direct application of this idea for economic growth is in Alesina and Tabellini (1989), which examines the effect of political uncertainty on investment and capital flight. The possibility of a government collapse leading to a new government prone to tax capital and productive activities implies a substitution of productive domestic investments in favor of consumption and capital flight, and thereby leads to a reduction of domestic production.
A different argument leading to a similar relation between political instability and growth is implied by Grossman's (1991) analysis of revolutions. In countries where rulers are relatively weak, i.e. more easily overthrown, the probability of revolutions is higher and the citizens have higher incentives to engage in revolutionary activities rather than productive market activities. On the contrary, a strong ruler who makes a revolution unlikely to succeed discourages revolutionary activities in favor of market activities.
A related line of research, in particular the work by Murphy, Shleifer and Vishny (1991) and Terrones (1990), emphasizes the negative effects of rent-seeking activities on economic growth. A weak government constantly under threat of losing office may be particularly sensitive to the need of pleasing lobbyists and pressure groups, thus leading to a more direct effect of rent-seeking activities on policy decisions.
Two objections to these arguments are worth mentioning. The first one is that a high propensity of a government change may be viewed favorably by economic agents if the current government is incompetent and/or corrupt and its possible successors are viewed as an improvement. In a large sample such as ours, it is reasonable to assume that the expected value of the competence of future governments is not higher than the current government competence.
Second, if the propensity of government change is large, an increase of it may actually reduce political uncertainty, since it becomes more certain that the current government will collapse. However, if the characteristics, or even the identity of the successor of the incumbent government are not known with certainty, an increase of the propensity of a political change may lead to an increase in policy uncertainty. In fact, it implies an increase of the propensity of substituting a well known (even though, possibly, inefficient) government for a less known one.
A study of the effects of political instability on economic growth needs to deal with the problem of joint endogeneity: even if it is true that a high propensity of having frequent government changes reduces growth, it may also be the case that low growth increases the probability of a government change. These are, in fact, the results shown by Londregan and Poole (1990), (199 la) in their studies of the economic determinants of unconstitutional transfers of power.
A related issue is whether democratic institutions are harmful or conducive to growth. A rather popular argument is that democratic institutions may be harmful to growth8 The basic idea underlying this view is that policy makers in democratic government are subject to the pressures of interests groups, and thus short-sightedly follow opportunistic policies to enhance their chances of reelection instead of 6 policies that enhance long term growth. However, these arguments against democracy are not necessarily conclusive. First of all, dictators may also need to be opportunistic if their survival in office is threatened. Second, authoritarian regimes are not a homogenous lot: they include "technocratic" dictators and "kleptocratic" ones. While the apparent association of high economic growth with authoritarian regimes is suggested by the experience of several authoritarian "technocratic" regimes (such as those in Korea, Taiwan, Indonesia, Turkey, Chile) it is as well evident that for each "benevolent" dictator one can observe at least as many "kleptocratic" and/or inept authoritarian regimes whose rule has led to systematic economic mismanagement and eventual political and economic collapse of their countries.' One can therefore conclude that, both on theoretical and empirical grounds, there is no obvious relationship between democracy and growth.
In fact, the empirical cross-country evidence on the relation between democracy and growth is quite mixed. Some early studies argue that democratic regime tend to slow economic growth while authoritarian regimes tend to stimulate it." However, others show that there is no systematic relation between long term growth and the democratic/ authoritarian nature of the political regime." Alesina and Rodrik (1991) present a model which is consistent with this inconclusive evidence. In their model, democracies should grow faster than "populists" or "kleptocratic" dictatorships, but less fast than "right wing" or "technocratic" dictatorships.

Methodology
This section describes the procedures employed for the estimation. First, we give a brief discussion of single equation estimation, where cross section growth regressions are considered. The primary purpose of employing this method is to facilitate a comparison of our results with those of other cross section studies in the recent literature such as those of Barro (1990Barro ( , 1991. A major drawback of a single equation approach for our study is that it does not take into account the joint endogeneity of the growth and government change.
Hence, later in this section, we turn to a discussion of a simultaneous equation methodology, which constitutes the primary focus of this study.

Single Equation Method
3.1 .a. Political Instability Political instability, defined as the propensity of an imminent government change, is not directly observable. Since "government change" is a discrete phenomenon, we employ limited dependent variable estimation methods. Propensity of government change is characterized as a function of economic and political variables. We estimate the probit specification described below using time-series cross-section pooled data (for notational convenience time and country indicators are omitted): where: 8 CL = a latent variable such that when c* > 0 we observe the occurrence of a government change, and we do not observe government change otherwise. x1 = variables (economic and political) that determine the occurrence of government change. q = normally distributed error term with mean zero. This specification facilitates an estimation of probabilities of government change that varies over time and across countries. We then average these annual measures of probability for each country over time so as to obtain a cross section measure of instability, which we call INS, to use in the cross section growth regressions described next.

.b. Economic Growth
A cross section estimation of growth is described with the following specification: Y= average economic growth in each country for the sample period.
x, = economic variables that explain economic growth, INS = measure of political instability, obtained from equation (1) as the average estimated probability of government change over the sample for each country. &= error term with mean zero.
This approach has two problems. First, as instability is a generated regressor, the standard errors of the second stage equation are generally inconsistent. l2 A more serious problem is that of simultaneity. Since the propensity of government change and economic performance are endogenous, equations (1) and (2) are both likely to be biased. We address this issue by using a simultaneous estimation of the two equations for growth and political instability as described next,

A Simultaneous Equations Approach
Let us define the following structural equation system, where the dependent variables of government change and growth are as before (but now both with yearly frequency): where: Y = annual rate of growth, X = exogenous variables that determine both government change and growth, x, = exogenous variables (economic and political) that determine the occurrence of government change only (i.e instruments for instability), xy = exogenous variables that determine economic growth only (i.e. instruments for growth), 49 u2 = error terms are assumed to be bivariate normally distributed with zero mean and variance covariance matrix that allows for cross-equation correlations.
The coefficients 7C and rY take into account the contemporaneous feedback between growth and changes of government, while the ar and 6 coefficients measure the effects of the exogenous variables. One way of identifying the system requires that at least one each of the & and X,, variables exist; that is, we need one exogenous variable in the growth equation which is not in the equation for government change, and vice-versa. An alternative way of identifying the system of equations is to impose restrictions on the contemporaneous feedback, i.e. yC=O or yy =O. In order to test the model (a chi-square test), there must at least be one overidentifying restriction, in addition to the restrictions needed to identify the model fully. We discuss the economic and political variables used as our identifying restrictions in Section 4. This model, a simultaneous equation system involving a latent variable, is described in Heckman (1978). While this system could in principle be estimated by standard maximum likelihood methods, the resulting likelihood function is extremely non-linear and thus difficult to maximize using standard methods. Londregan and Poole (1990) use the results of Newey (1987) to estimate this type of system through an application of Amemiya's Generalized Least Squares Technique (AGLS). Since we employ the same econometric methodology the technical details are not replicated here (see the Appendix of Londregan and Poole (1990)). Instead, we provide only a heuristic description of the estimation procedure. The reduced forms take into account that there may be shocks common to both growth and instability.
The structural form estimates take into account the simultaneous feedback between growth and government change. Fully efficient structural estimates are obtained by a GLS regression of the stacked coefficients from the two reduced form equations against the two "contemporaneous" (y) parameters and a "selection" matrix which picks out the appropriate reduced form coefftcients. A bootstrapped estimate of the variance-covariance matrix of the reduced form coefficients (Efron (1982)) is used to form the weighting matrix for the GLS regression. We use 1024 bootstrap replications (so that the number of replications we have are identical to Londregan and Poole (1990)).

Data and Specification
This section briefly describes our data and the specification of our equations for political instability and for growth. Our panel data set includes a time series and cross section panel for 113 countries.13 For about half of the countries the sample period is 1950-82, for most of the others the sample is 196042. A list of countries and sample period is in Table A. 1, in the Appendix.

The Specification of the Political Instability Equation
Our specification for the government change is similar to those employed in the recent literature (Cukierman et al. (1992), ozler and ). The independent variables can be classified in three broad classes: 1) indicators of political unrest such as cabinet adjustments; 2) "structural" institutional variables which account for differences across countries such as the GDP per capita and being a democracy or not; 3) economic performance in the recent past, in particular the recent growth level. A complete list, along with definitions and sources of each variable is provided in Table A-2. A significant innovation in our data concerns the definition of the dependent variable for government change. We employ two different dependent variables. The first one (GCHANGE) is the one used in previous studies and obtained from Jodice and Taylor (1983).
This variable codes as one any regular or irregular (i.e., coup) transfer of executive power.14 In an attempt to eliminate from our dependent variable government changes which do not involve a substantial turnover of 13 leadership, we have constructed a variable for major changes (MJCHANGE). This includes: i) all "irregular" transfers of power; ii) a subset of "regular" transfers which imply a change in the party, or coalition of parties in office. This change in the definition substantially reduces the number of "changes" in our dependent variable. The sample characteristics of our data are displayed in Table   1.
A second innovation in our data set is our own construction of a variable for democratic institutions, DEMOC. This variable takes the value of one for countries with free competitive general elections with more than one party running; two for countries with some form of elections but with severe limits in the competitiveness of such ballots; three for countries in which their leaders are not elected.15

The Specification of the Growth Equation
The variables employed are described in Table A.2, separately for the cross-section data used in the single equation estimation, and the panel data used in the simultaneous equation estimation (differences between the two primarily arise from data availability).
Our specification draws heavily upon the recent growth literature. We include variables which proxy for the level of income and the level of human capital, as well as regional dummy variables.
In the time series cross section specification we also control for the world business cycle by adding the weighted average of the lagged growth rate of the seven largest industrial economies and we control for "persistence" in growth by using the lagged dependent variable.
In the joint estimation we use both economic and political variables to identify the growth and government change equations. In our base specification, the growth equation is identified by the enrollment ratio in primary school (EDUC). The government change equation is identified by the lagged dependent variable and by a dummy variable that indicates the occurrence of an executive adjustment (EXADJ), lagged one period. These three identifying assumptions imply one overidentifying restriction which can be tested. The test has a chi squared distribution with one degree of freedom; it essentially tests the difference between the reduced and structural form. A small value for the test statistic corresponds to a high p-value, which indicates the significance level of not rejecting the model.

Results of the Single Equation Approach
This section presents cross-sectional results which extend previous work by Barro (1989Barro ( , 1991, Scully (1988), and Kormendi and Mcguire (1985). The joint endogeneity issue is addressed in the next section, where we estimate a simultaneous system; here, we first derive a measure of political instability and use it in cross-section growth regressions. As specified in (1) and (2) above, our procedure for constructing a measure of the probability of government change is to estimate a probit model of total government change on pooled time series and cross-country data. We then use the fitted values from this probit regression to derive the average predicted probability of a government change over the sample period for each country. The results of these probit regressions are presented in Table A.3 in the   15 Appendix. l6 The next step is to introduce our estimated measure of political instability from the probit regression in standard crosssectional regressions of the determinants of economic growth.
Before presenting these results, we show in column (1) of Table   2 a replication of Barro's regression for the average per capita growth rate of the 98 countries in the sample in the 1960-1985 period (for a list of countries see Table A-l). The results of this regression are familiar. Initial per capita income (GDP60) has a significant negative sign, confirming the hypothesis of conditional convergence; high economic growth is associated with high initial level of human capital (PRIM60 and SEC60); non-productive government spending (GOV) and distortions in the pricing of capital goods (PPSODEV) lead to lower economic growth; and the two regional dummies for Latin America and Africa are significant with a negative sign. Finally, the coefficient on REVCOUP is negative and significant, indicating that violent government changes are associated with lower growth, while the assassination variable (ASSASS) has the right sign but is not statistically significant.
In column (2) of Table 2 we replace the REVCOUP and ASSASS variables with our measure of political instability (INS); this is the average predicted probability of a government change. The coefficient estimate is negative and significant at the 1 percent confidence level; after controlling for the other determinants of growth, per capita growth is lower in countries characterized by a higher degree of political instability. Column (3) shows that our measure of political instability remains significant even when the REVCOUP variable is included. Similar results (nut displayed in the Table) are obtained when the ASSASS variable is also included.
Column (4) of Table 2 shows that a dummy for "democratic" regimes, DEMOCAV, is insignificant. Column (5) reports our results using a measure of political instability (MJINS) in which we consider only major changes of government. Thus, the dependent variable in the probit regression (in table A.4) used to obtain MJINS is

MICHANGE.
This measure of instability is also significant.
Furthermore, the coefficient is more than double in absolute value than that of the variable INS in column (2). As expected, major government changes have a more serious effect on growth.
Column (6) of Table 2  In summary, this section has shown that the degree of political instability, as proxied by the probability of government change negatively affects per capita GDP growth. However, these singleequation regressions do not properly take into consideration the problem of joint endogeneity between growth and political instability.
This issue is examined in the next section.  The corresponding structural form estimates are presented in Table 4. Inspection of these results suggests the following findings concerning the contemporaneous feedback effects: 1) The impact of political instability on growth, captured by the coefftcient on  changes" on growth remains significant. The coefficient on the effect of growth on the propensity to observe major government changes is negative, but not statistically significant (though the "t" value reaches -1.38 in the the large sample).

Redts of the Simultaneous Equations Approach
In contrast to the results in Table 4, the coefficient on the Latin America dummy variable in the government change equation is positive and significant. This is not a surprising result once it is noted in Table   1 that while the frequency of GCHANGE for Latin America is actually lower than in the industrial and Asian countries, the frequency of MICHANGE in Latin America is the highest in the world, almost double that in Asia and a third larger than in the industrial countries.
These figures highlight how Latin America is a region with frequent major political changes and, as emphasized above, with low growth.
This result supports the idea that what is particularly harmful to growth 20 is polarization in the society and in the political arena leading to frequent substantial turnovers of political direction.
It is also worth emphasizing how it is likely to be the case that various events concerning political unrest such as government changes, attempted coups and executive adjustment are likely to be underreported in African countries. Note that Table 1 highlights how Africa has fewer government changes and executive adjustments than any other region in the world. Our result concerning the effect of political instability on growth would probably be strengthened by a correction of this underreporting bias. In fact, Africa is a region with low growth and an underestimated measure of political instability.

Sensitivity Analysis and Discussion
Our basic result that political instability is harmful to growth is robust to changes in the model specification discussed below.
No significant changes in the results are found when we add additional lags of EXADJUST and when we introduce the variable ATTEMPT, which measures unsuccessful attempts to change the government, including aborted coups, into the government change equations. We also considered an industrialized country dummy INDUST for the growth equation.
Specifications that incorporate the level of GDP (either GDP60 or GDPl, i.e., lagged level of GDP) are considered. The effect of political instability on growth remains statistically significant in all these specifications. However, the chi-square statistics are much larger than the ones found earlier, leading to a rejection of the model.

The effect of democratic institutions is investigated by adding
the variable DEMOC in the growth equation. This specification of the model is rejected based on chi-squared tests, and the variable DEMOC is not found statistically significant.
As an alternative means of identifying the model the contemporaneous effects of growth on government changes is set equal to zero (note from Tables 3 and 4 that this parameter is not found statistically significant). The model is not rejected, but gives a much lower level of significance in comparison to specifications that do not impose this restriction. A consequence of this restriction is that coefficient on lagged GDP growth in the government change and major change equations becomes significant, indicating that low lagged growth increases the propensity to government changes.
Finally, Table 6 presents our base specification for the case in which the dependent variable for government changes is coup d'etat, as in Londregan and Poole (1990). We provide this result as a way of suggesting where the differences between the two sets of works might be arising form.
Unlike that study, we continue to find a negative and a statistically significant effect of the propensity of government change (now a coup detat) on growth. By focusing only on coup detat we in effect reduce the primary difference between these two sets of works to model specification (our general data base and econometric methodology and now the government change variable are almost identical An interesting finding that emerges in Table 6 is that the contemporaneous growth has a significant impact on propensity of a coup detat. This result may indicate that when growth is low, one tends to observe "substantial" changes of political control rather than more reappointments of incumbents. This of course needs further investigation, as not all coup detats are substantial if it only involves a turnover of generals that is not associated with major political or economic uncertainty in the country.

Concluding Remarks
Political instability reduces growth. This finding is very robust: it has been obtained in a model in which several other economic determinants and "regional" factors affecting growth and political stability are accounted for. Democracies do not appear to show a different growth performance than non-democracies. Also, the occurrence of a government change increases the likelihood of subsequent changes, suggesting that political instability tends to be persistent.
Rather than reviewing in detail all our results, we conclude by highlighting several possible extensions of this paper. First, it is 23 worthwhile to continue in our effort to distinguish cases of "major" government changes from "routine" turnovers of leadership with no significant changes in the ideological orientation of governments. Our efforts have been, thus far, reasonably successful in the sense that the results using our new variable improve, on some grounds, relative to those obtained with the Jodice and Taylor (1983)  Second, one may try to classify the "ideological" direction of government changes and test for the effects of different government's ideology on economic growth. Such a classification on a "left" and "right" scale is reasonably easy for a subset of countries (for instance, OECD democracies), but much more problematic for other countries where religious, tribal or regional conflicts dominate the polity." An even more difficult but useful approach would be to attempt to measure the degree of ideological polarization of various parties and groups in different countries.
A third extension would be to include measures of income distribution. Alesina and Rodrik (1991) and Persson and Tabellini 24 (1991) have found a negative relationship between income inequality and growth, particularly in democracies. Political unrest may also be influenced by inequality while economic development is bound to affect income distribution. Thus, the three variables, growth, political instability and income distribution are jointly endogenous. The difficulty in pursuing this extension is mostly due to the availability of reliable time series data on income distribution.
A fourth direction of research would be to pursue further the analysis of democracy and growth. As emphasized above, dictatorships are far from homogeneous. It would be quite interesting to engage in a disaggregate analysis of which politico-institutional characteristic of dictatorships make them more or less growth-enhancing. Some recent results by Poole and Londregan (1991b) suggest that this may in fact be a promising avenue of research. In a reduced form growth equation they find that the presence of "unconstitutional leaders" reduces growth. Their coding of "unconstitutional leaders" is probably close to capturing a subset of truly "kleptocratic" dictators.
A fifth direction of research is to study a "non linear" pattern of influence from economic growth to political stability and vice-versa.
For instance, Huntington (1968) suggests that while at relatively a high level of income political stability and growth go hand in hand, periods of "take off", i.e., of exceptionally high growth, in middle-income countries may be associated with rapid social transformation and political unrest. More generally, the interaction between political stability, political change and growth may take different forms at different levels of development.  Edwards and Tabellini (1990), Edwards and Tabellini (1991), dzler and Tabellini (1991), Roubini (1991), Sachs (1989a, 1989b

This "incompatibilityhypothesis"
has been formulated in a number of ways: see Huntington andNelson (1976), Kahn (1979) and Olson (1982 (1974( , Goldsmith (1987 and the recent systematic work by Weede (1983)        The date next to each country in this group indicates the beginning of the sample for which data are available.