Political Instability and Economic Growth

Transcription

1 Journal of Economic Growth, 1: (June, 1996) Kluwer Academic Publishers, Boston. Political Instability and Economic Growth ALBERTO ALESINA Department of Economics, Harvard University, Cambridge, MA 02138, National Bureau of Economic Research, and Center for Economic Policy Research SULE OZLER Department of Economics, University of California at Los Angeles, Los Angeles CA NOURIEL ROUBINI Stern School of Business, New York University, New York, NY 10012; National Bureau of Economic Research; and Center for Economic Policy Research PHILLIP SWAGEL Department of Economics, Northwestern University, Evanston, IL This paper investigates the relationship between political instability and per capita GDP growth in a sample of 113 countries for the period 1950 through We define political instability as the propensity of a government collapse, and we estimate a model in which such a measure of 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. We also discuss the effects of different types of government changes on growth. Keywords: political instability, economic growth, government changes, coup d'etat JEL classification: Introduction At the beginning on this century Argentina was one of the wealthiest countries in the world. In 1960, Argentina's income per capita was in the top twenty in the world and was higher than that of Japan. In the last thirty years, however, Argentina has often come close to economic collapse. In 1960, Japan had a per capita income below Iraq, Ireland, and Argentina and was not even in the top twenty-five in the world. Since then Japan has experienced one of the fastest growth rates in the world. Argentina has had a history of political instability, with several coups d'etat and much political violence. In contrast, until very recently Japan has been a model of political stability, with the same political party in office continuously from 1960 until This paper addresses two questions: Does this tale about Argentina and Japan capture a more general correlation between economic growth and political stability? Does political

2 190 ALESINA, I~ZLER, ROUBINI, AND SWAGEL stability foster economic growth, or does low economic growth lead to political instability? On the one hand, the uncertainty associated with an unstable political environment may reduce private investment and, therefore, growth. 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 collapse (our measure of "political instability") and growth of per capita GDP in a sample of 113 countries for the period 1950 through After controlling for several economic determinants of growth, we find that in countries and time periods with a high degree of political instability, growth is significantly lower than otherwise. The effect of growth on political instability is less clear: the effect of low economic growth on government collapses is strong for coups d'etat but much less clear for other types of government change. We also find that political instability is persistent: the occurrence of government changes in the recent past increases the probability of observing future collapses. Finally, we do not detect any difference in the average rate of growth between authoritarian regimes and democracies, even after controlling for other political factors. Three important features characterize our study. First, we focus on government changes as an indicator of political instability; we classify changes depending on whether or not they are constitutional and whether or not they involve significant modifications in the ideological orientation of the executive. Second, we explicitly allow for joint endogeneity of growth and government changes. Third, we account for the fact that what matters for economic activity is not only the actual occurrence of a government change but also the expectation of it. In their studies of military coups, Londregan and Poole (1990, 1991a) have dealt with similar problems of joint endogeneity by estimating a system of equations in which the dependent variables are growth and coups d'etat. 2 Our model differs from theirs in several important respects. First, our dependent variable for government changes is different. Unlike them, we do not focus solely on coups but instead examine broader definitions of government change. Second, our model is substantially different from theirs. In particular, unlike them, in the growth equation we control for several economic factors identified by the recent empirical literature on endogenous growth. Because of these differences, our results are quite different from theirs, particularly on an important point: we find robust evidence of a negative effect of political instability on growth while they do not. A related literature has studied the effect of political instability on growth using various measures of instability. Hibbs (1973), Gupta (1990), and Alesina and Perotti (1996), for example, measure political instability by constructing indices which summarize data on the occurrence of political violence and unrest. Barro (1996), Benhabib and Spiegel (1994), and Easterly and Rebelo (1993), among others, add indicators of political instability to cross-section regressions in which the dependent variable is either growth or investment. These papers generally find that political instability reduces growth or investment. None of these papers, however, takes into account the joint endogeneity of growth and political instability, nor do they explicitly take into account the effect of instability on expectations. 3 Cukierman, Edwards, and Tabellini (1992), Roubini (1991), and Ozler and Tabellini (1991) adopt a measure of political instability similar to ours in their studies on inflation, budget

3 POLITICAL INSTABILITY AND ECONOMIC GROWTH 191 defecits, and external debt. While they find the expected effects in cross-section regressions, they do not take into account problems of reverse causality. This paper is organized as follows. Section 2 discusses the basic questions that we explore in our empirical analysis. Section 3 describes our data set and highlights some basic statistics. Section 4 presents the econometric methodology, after which Section 5 describes our empirical specification. In Section 6 we present our basic results, with sensitivity analysis in Section 7. The last section discusses some possible extensions. 2. Political Instability and Economic Growth: Theoretical Issues 2.1. A Theoretical Framework We define political instability as the propensity of a change in the executive power, either by constitutional or unconstitutional means. We study whether a high propensity of an executive collapse leads to slower growth and, conversely, whether low growth increases the propensity of a government change. Political instability affects growth because it increases policy uncertainty, which has negative effects on productive economic decisions such as investment and saving. A high probability of a change of government implies uncertain future policies, so that risk-averse economic agents may wait to take productive economic initiatives or might even "exit" the economy by investing abroad. Similarly, foreign investors are likely to prefer a stable political environment. 4 On the other hand, low growth increases government instability. A vast empirical literature has shown that in industrial democracies incumbent governments' chances of reelection depend on the rate of growth immediately before the elections. In nondemocracies, low growth increases popular dissatisfaction, creates incentives for antigovernment activities, and may make coups d'etat more likely. 5 The interaction between growth and political instability can lead to a vicious circle: suppose that for some exogenous reason the probability of a government collapse increases. This might result, for example, from an increase in political conflict unrelated to the economy or from international political developments. Investment and growth fall as a result of the shock, further increasing the likelihood of a government collapse, leading to even more political uncertainty. On the other hand, suppose that the rate of growth falls for some exogenous reason--for instance, an adverse movement in a nation's terms of trade. The public will hold the government responsible, at least in part, for the poor economic outcome.6 This increases the probability of an executive collapse, reducing growth even more. The effect of uncertainty about government changes is larger the higher the degree of political polarization, where polarization is defined as the difference in economic preferences between different groups and political parties. The probability of a government change may not have much effect on the expectations of future economic policies if the next government is likely to follow policies similar to those of its predecessors; in highly polarized societies, instead, government changes may lead to radical changes in policy making. Thus, in our empirical work we distinguish between different types of government changes, in order to isolate those that are more likely to imply significant changes in economic policies.

4 192 ALESINA, OZLER, ROUBINI, AND SWAGEL Before proceeding to our empirical analysis, we must tackle three possible objections to our hypothesis as to the effect of political uncertainty on growth. The first one is that a high probability of a government change might be viewed favorably by economic agents if the current government is incompetent or corrupt and its possible successors are seen as an improvement. In our large sample of countries, it is reasonable to assume that government competence is a random walk, in the sense that the expected competence of future governments is not higher than the current government competence. The hypothesis that we test, then, is that the negative effect of political uncertainty dominates the effect of a possible improvement in the quality of the government. Second, if the probability of a government change is above one-half, an increase in this probability might actually reduce political uncertainty, since it becomes more and more certain that the current government will collapse. However, this might still lead to an increase in policy uncertainty if the characteristics and even the identity of the successor government are not known. Third, since we emphasize policy uncertainty, why don't we focus on economic policy directly rather than on government changes? 7 In different countries and time periods the relevant policy instruments might vary: perhaps monetary policy might reflect political uncertainty in one country, while fiscal policy, taxation, or changes in trade policy might be important in others. In a large sample such as ours, then, it would be difficult to focus on a small group of policy variables Related Theoretical Approaches While we have justified our hypothesis on the effect of instability on growth based upon the effect of policy uncertainty, Grossman's (1991) analysis of revolutions leads to a similar correlation. In countries where rulers are relatively weak--that is, more easily overthrown-- the probability of revolutions is higher, and citizens have larger incentives to engage in revolutionary activities rather than productive market activities. Thus a strong ruler who makes a revolution unlikely to succeed might act to discourage revolutionary activities in favor of market activities. Murphy, Shleifer, and Vishny (1991) and Terrones (1989) emphasize 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 to please lobbyists and pressure groups; this might lead to a more direct effect of rent-seeking activities on policy decisions. Furthermore, Shleifer and Vishny (1993) argue that "weak" governments exhibit a type of corruption that is more deleterious to economic growth than relatively corrupt but "strong" governments. On the other hand, several authors have argued that political stability and growth may not always go hand in hand. For example, Olson (1982) emphasizes that governments that remain in office for a long duration become easier prey for interest groups and are thus more likely to follow policies that do not maximize social welfare. Huntington (1968) contends that periods of "take off" and rapid accumulation create social disruption and political unrest.

5 POLITICAL INSTABILITY AND ECONOMIC GROWTH Democracy and Growth A related issue is whether democratic institutions are harmful or conducive to growth. In a democracy, defined as political system with free competitive elections with at least two parties represented, policy makers are subject to the pressures of interest groups and might short-sightedly follow opportunistic policies to enhance their chances of reelection rather than policies that enhance long-term growth. Dictators may well be less sensitive to these problems. These arguments are not 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 and so on), it is as well evident that for each benevolent dictator, one can observe at least as many kleptocratic or inept authoritarian regimes whose rule led to systematic economic mismanagement and eventual political and economic collapse of their countries. One can therefore conclude on both theoretical and on empirical grounds that there is no obvious relationship between democracy and growth. 8 This observation justifies our inclusion of both democracies and dictatorships in the same sample. In any case, we will test for systematic difference in our results among the two groups. 3. Data and Some Basic Statistics Our data include a panel of annual observations for 113 countries. For about half of the countries the sample period is 1950 through 1982, while for most of the others the sample is 1960 through Appendix A contains a list of countries and sample periods (Table A. 1), along with detailed definitions and sources for each of our variables (Table A.2). Our sample does not include former Soviet bloc countries. The source for our economic data is Summers and Heston (199l), while most of our political data come from Jodice and Taylor (1983) and Banks (various issues). A significant innovation in our data concerns the definition of the dependent variable for government change. We employ three different variables. The first definition, GCHANGE, is obtained from Jodice and Taylor (1983). 9 This variable codes as a 1 a year in which there is any regular or irregular (that is, coup) transfer of executive power, with a zero otherwise. In an attempt to eliminate from our dependent variable government changes that do not involve substantial turnover of leadership, we have constructed a second variable for major changes, MJCHANGE. This includes all irregular transfers ofpower such as coups, along with the subset of regular transfers that imply a substantial change in the party or coalition or parties in office. This restricted definition of government turnover greatly reduces the number of changes in our dependent variable. 1~ We have constructed our new variable, MJCHANGE, to eliminate from the definition of a government change those executive turnovers that, although formally implying a change in the executive, do not bring about a change in the political orientation of the leadership.l 1 Finally, we also use a third dependent

6 194 ALESINA, 0ZLER, ROUBINI, AND SWAGEL variable, C O U P, which includes only irregular transfers of power such as military coups. This is also obtained from Jodice and Taylor (1983). Another innovation in our data set is that we constructed a variable for democratic institutions, DEMOC. This takes the value of 1 for countries with free competitive general elections with more than one party running, 2 for countries with some form of elections but with severe limits in the competitiveness of such ballots, and 3 for countries in which leaders are not elected, lz Table 1 describes the basic features of the data. 13 The average frequency of total government changes (GCHANGE) for the sample of all countries is 0.28; that is, we observe a government change on average about every three years. The frequency of major government changes (MJCHANGE) is 0.11, while the frequency of irregular government changes (COUP) is about Total government changes are most frequent in the industrial countries (0.39) and least likely in Africa (0.20); military coups are most frequent in Latin America (0.079) and Africa (0.060), and nonexistent in industrial countries. While government changes are frequent in industrial democracies (0.39), major government changes are much less frequent (0.12), suggesting that most governments are followed by others with a similar orientation. Latin America is interesting since it has a frequency of total government changes about equal to world average (0.29) but has the highest frequency of both major government changes (0.16) and of military coups (0.079). Thus, these data highlight the well-known political instability of this region. In Africa, changes in the government are rare, particularly regular changes. African countries are typically authoritarian regimes with few regular elections and few alternations in power. In this region, executive changes take the form of major changes (0.10), out of which military coups are more than half (0.06). In Asia, government changes are close to the world average (0.30), but major government changes are much more rare than in any other region. Asia has the second-lowest frequency of coups (after the industrial countries), while the democracy variable for Asia averages 2.32, close to the authoritarian index of 3. These data confirm the view of Asia as a region with authoritarian but stable political regimes. Finally, the "Other" region includes six European countries that are economically less advanced than the rest of Europe and are characterized by unstable political histories and, at times, authoritarian regimes: these are Greece, Malta, Portugal, Turkey, Yugoslavia, and Cyprus. In these countries, the frequency of major government changes and coups is about as high as in Latin America. In summary, the two regions with the lowest growth rates (Latin America and Africa) are also the two most unstable regions. Also, this table does not show a correlation between democracy and growth. Table 2 presents averages of growth rates separately for years with and without government changes. The first column considers the sample of all countries. The average per capita growth rate in a year without a government change is 2.8 percent; growth falls to about half this value in years with a government change. The difference is even stronger if we consider only major government changes: growth averages only 0.1 percent in years with a major government change. Finally, the growth gap is most striking for coups: the average growth rate in a country year in which a military coup occurred is -1.3 percent. 14 The fourth line reports the average growth rate in years with government changes that are not

7 POLITICAL INSTABILITY AND ECONOMIC GROWTH 195 Table 1. Sample means of the data: (standard deviations in parentheses) Latin Industrialized All America Africa Asia Countries Other GCHANGE (.45) (.45) (.40) (.46) (.49) (.48) MJCHANGE (.32) (.36) (.30) (.25) (.32) (.37) COUP (.21) (.27) (.24) (.19) (.00) (.24) EXADJ (.50) (.50) (.50) (.50) (.50) (.50) DEMOC (.93) (.92) (.50) (.89) (.37) (.91) GROWTH (.069) (.065) (.085) (.068) (.035) (.060) WGROWTH (.019) EDUC (.29) (.17) (.32) (.24) (.16) (.15) GDP (4119) (1004) (508) (8316) (1801) (165) GDP (3822) (1622) (893) (5362) (2484) (1149) Countries Observations ( ) Observations ( ) Note: Region breakdowns use the IMF coding system. Hence, the "other" category refers to the nonindustrialized European countries. major. This growth rate is only slightly below the growth rate of years with no changes. This observation suggests that it is the major government changes that are most strongly associated with low growth. The other columns of Table 2 show that these observations are generally common to every region of the world. In every region, growth is highest in years with no change and lower in years with government changes. As before, years with major changes exhibit worse growth performance than years with just regular changes, while years with coups have the lowest growth rate. The common features of these data support our pooling together of data for every region. Note, however, that in years with coups, growth is substantially higher in Asia than in coup years in every other region of the world. Table 3 presents essentially the same data as in Tables 1 and 2, but ranks the countries by level of per capita income. The most important observation for our purpose is that the comparison between the rate of growth in years with no changes and the three different types of government changes is common to all the income groups. Leaving aside coups in the richest quartile (we have only two observations in this category), in every income quartile growth is highest in years with no change, lower with GCHANGE, even lower with MJCHANGE, and lowest in years with COUPS. These observations suggest that the pattern of correlation between executive changes and growth is not specific to a certain

8 196 ALESINA, OZLER, ROUBINI, AND SWAGEL Table 2. Government change and growth: average per capita growth rates in country years with and without a government change. Latin Industrialized All America Africa Countries Asia Other Years with NO GCHANGE Number of observations Years with GCHANGE Number of observations Years with MJCHANGE Number of observations Years with COUP Number of observations Years with GCHANGE but not MJCHANGE Number of observations Note: Sample is 1960 through Table 3. Growth and government changes by income groups Income Quartile (poorest) (riches0 Average GDP ($1,985) 696 1,487 3,220 8,604 Average GROWTH all years Average GROWTH with NO CHANGE Average GROWTH with GCHANGE (all changes) Average GROWTH with MJCHANGE (major changes) Average GROWTH with COUP Average GROWTH with GCHANGE only (not major changes) Note: Sample 1960 through income level. Note, however, that in the richest quartile the difference between years with MJCHANGE and no change is the smallest. It is also interesting to note that the level of political instability, judged by the frequency of M J C HANG E and C O U P S is highest in the second quartile rather than in the poorest one.

9 POLITICAL INSTABILITY AND ECONOMIC GROWTH 197 The differences in the growth rate between country years with and without government changes are generally very significant (based on formal F-tests) both across the three definitions of government change and for different regional subgroups and income groups. However, these observations alone are not sufficient to identify the direction of causality between growth and government change. Is the observed correlation in the data explained by political instability causing low growth, or is it instead that low growth leads to government collapse and government change? To answer this question we must examine the joint determination of government change and economic growth. We next describe the methodology and our empirical specification of such a model. 4. Methodology 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. The propensity of government change is characterized as a function of economic and political variables. A single equation approach has a problem of simultaneity: since the propensity of government change and economic performance are endogenous, estimates of the effects of instability on growth without controlling for the effects of growth on instability will produce biased estimates. We address this issue by using a simultaneous estimation of the two equations for growth and political instability. Define the following structural equation system, where the dependent variables are annual observations of government change and growth, and where time subscripts are omitted: where c* = ctcxc + flcx + YcY + Uc y = OlyXy -]- j~yx -]- yyc* --[- Uy, (1) c* = a latent variable such that when c* > 0 we observe a government change, and c* < 0 we do not observe a change, 15 y = yearly rate of growth, X = exogenous variables that determine both government change and growth, Xc = exogenous variables that determine the occurrence of government change only, Xy = exogenous variables that determine economic growth only, and uc, Uy = bivariate normal errors. The coefficients ~,c and yy take into account the contemporaneous effect between growth and changes of government, while the ct and/3 coefficients measure the effects of exogenous variables. One way of identifying the system requires that at least one each of the Xc and

10 198 ALESINA, 13ZLER, ROUBINI, AND SWAGEL Xy variables exist; that is, we need one exogenous variable in the growth equation that is not in the equation for government change, and vice versa. 16 Because one of the endogenous variables is a latent variable, standard econometric methods for simultaneous equations systems such as two-stage least squares cannot be applied. Newey (1987) shows how to estimate this type of system through an application of Amemiya's generalized least squares technique (AGLS) (Amemiya, 1978). A description of this estimation method is available on request and is available in the working paper version of this paper. 5. Model Specification The explanatory variables in our government change equation can be classified in three broad classes: (1) indicators of political unrest that can explain the imminence of a government collapse, such as recent executive adjustments and recent attempts to overthrow the government, (2) structural variables that account for differences across countries such as being a democracy or belonging to a geographic region, and (3) economic performance in the recent past, in particular the recent growth rate. Our basic specification of the government change equation includes the contemporaneous growth rate (G R O W T H), lagged growth (G R O W T H (- 1)), and lagged world growth (WG R 0 WTH (- 1)); this last variable is the GDP-weighted growth rate of the G-7 nations and is included in the government change equation to examine whether economic performance relative to the rest of the world is relevant for the popularity of a government. 17 We also include Latin American and African regional dummy variables (LATIN and AFRICA) to capture the institutional features (the electoral laws, history of democracy, and authoritarianism) of these regions, the lag of the government change indicator variable (GCHANGE(-1)) in order to test for the persistence of political instability, and lagged executive adjustments (EXA DJ (-1)) as an indicator of incipient politicial unrest. The variable for executive adjustment is the number of changes in the executive branch that do not result in a change of leadership; this might be thought of as the number of cabinet reshuffles. We use other variables as measures of political unrest in the alternative specifications discussed in Section 7. Our specification of the growth equation draws on the recent growth literature such as Barro (1991). Our basic specification includes the following variables: the contemporaneous government change propensity to test for the causality between growth and instability, the enrollment rate in primary education (ED UC) as a proxy for the level of human capital, and Latin American and African regional dummies (LATIN and AFRICA) to control for region-specific factors that affect long-term growth rates. These regional dummies are generally significant in cross-section growth regressions (Barro, 1991). We include two additional regressors to control for the temporal variation in growth: the lagged growth rate (G R O W T H (- 1)) captures the persistence of the growth process, while the lagged world growth rate (WGROWTH(-1)) captures the effect of the world business cycle on the growth rate of a country.

11 POLITICAL INSTABILITY AND ECONOMIC GROWTH 199 and Our basic two-equation system is thus Prob(GCHANGE) = r + VtclEXADJ_I + Otc2GCHANGE_I +flco + flclg RO WT H-1 + flc2 WG RO WT H_I +flc3latin + ~c4africa), (2) GROWTH = yygchange + OtylEDUC q- ~yo q- flylgrowth-1 -k-fly2wgrowth_i -I- ~y3latin + ~y4africa -I- Uy. (3) The same specification as in (2) and (3) is also adopted for the systems in which the dependent variables are major changes, MJCHANGE, and coups CO UP, with the corresponding lagged dependent variable instead of the lagged of GCHANGE. TM Note that we do not include the level of per capita GDP in our basic specification; we discuss this point in Section 7. As shown in equations (2) and (3), we use both economic and political variables to identify the growth and government change equations. The variable EDUC (enrollment in primary school) enters the growth equation only, while lags of executive adjustments and government changes (the dependent variable) enter only in the government change equation. Previous research has shown that ED U C is an important determinant of growth, while it seems unlikely that this would directly affect the probability of a government change. Similarly, executive adjustments and lagged government changes are assumed to influence growth only indirectly and are thus excluded form the growth equation. The motivation for this choice of identification is that lagged occurrence of political turmoil (government changes and cabinet reshuffling) affect growth not directly but through their influence on the perception of further political instability--that is, future government change. These three restrictions (one in the equation (2), two in equation (3)) leave us with one testable overidentifying restriction; we discuss these tests in the next section. Needless to say, the choice of identifying restrictions is always open to criticism. Therefore, in Section 7 we discuss sensitivity tests on the model specification. 6. Results of the Joint Estimation Table 4 displays structural estimation results of equations (2) and (3) for two samples: the large sample, covering the 1950 through 1982 period, and a smaller 1960 through 1982 sample. While the large sample includes more information, pre-1960 data are not available for most African countries and several Middle-Eastern and Latin American countries (see Appendix A, Table A. 1). We report both estimates to emphasize that our results are robust to the sample choice. The impact of political instability on growth, captured by the coefficient on GCHANGE in the growth equation, is negative in both samples. This coefficient is quite significant in the large sample (1 percent level) and statistically significant at a lower level of confidence in the small sample (better than 10 percent level). The value of this coefficient implies that a change in the propensity of government change from zero to 1 brings about a fall

12 200 ALESINA, OZLER, ROUBINI, AND SWAGEL Table 4. Joint estimation of growth and government changes: base specification (t-statistics in parentheses) GROWTH GCHANGE Sample Sample Government Government Growth Change Growth Change Equation Equation Equation Equation (1) (2) (3) (4) (0.30) (-0.02) (-2.79) (--1.72) CONSTANT (-0.48) (--8.34) (--0.97) (--4.91) GROWTH(-1) (0.53) (-0.50) (2.93) (-1.11) W G R O W T H (- 1 ) (6.19) (0.11) (6.71) (0.41) LATIN (-4.37) (-1.04) (-3.64) (-1.63) AFRICA (-2.16) (-1.85) (-2.95) (-5.45) EDUC (2.28) (2.33) EXADJ(-1) (4.65) (4.76) GCHANGE(--1) (3.59) (4.35) X 2 test of the 1 overidentifying restruction p-value Number of observations in the growth rate of 1.3 to 1.4 percent per year. In terms of standard deviations, a one standard deviation increase in the propensity to government change decreases growth by about one-tenth of its standard deviation.19 On the other hand, current low growth does not appear to increase the propensity of a government change; this can be seen in the coefficients for the GROWTH variable in columns 2 and 4: these are small and statistically insignificant for both samples. These results suggest that the observed correlation between low growth and a high frequency of government change is the consequence of political instability causing low growth rather than low growth leading to a higher probability of government collapse. Let us now turn to the other determinants of growth. The coefficients on the econmic variables in the growth equation have the expected signs and are generally statistically significant. Our proxy for lagged world growth (WGR 0 WTH(-1)) is significant in both samples and captures the effect of a world business cycle; on average 1 percent increase in the growth rate of the G-7 countries increases growth by about one-half of a percent. The level of education has a positive and significant influence on growth; the result is consistent (and of the same order of magnitude) with cross-sectional work on growth such as Barro

13 POLITICAL INSTABILITY AND ECONOMIC GROWTH 201 (1991). The coefficient on lagged growth is statistically significant and positive in the small sample indicating persistence. The coefficient is positive but statistically insignificant in the large sample. The results for the government change equation are also quite sensible. The occurrence of executive adjustments in the previous year (EXADJ(-1)) significantly increases the propensity of a government change. Evaluating the government change equation (3) at the mean values for developed countries and assuming that there was not a governmnet change in the previous year, we find that changing EXADJ(-1) from zero to 1 increases the probability of a government change from to o The signific ant coefficient on the lagged dependent variable, G C H A N G E (- 1), shows that government changes tend to be persistent. Evaluating at the mean values for developed countries and assuming again that there were no executive adjustments the previous year, we find that a government change the previous year (GCHANGE of 1 instead of zero) increases the probability of a change this year from to Finally, low growth in the recent past does not increase the propensity of a government change: the coefficient on GR 0 WTH(-1) in the government change equation is negative but insignificant. The chi-squared tests of our overidentifying restriction shows that in the large sample our specification is not rejected at the 0.65 level of significance, while in the small sample the model is not rejected at the 0.80 level of significance. 21 Table 5 displays the specifications in which the dependent variable is major government changes, MJCHANGE, instead of GCHANGE. The coefficients of MJCHANGE on growth are negative and statistically significant in both samples, though as before our results are somewhat stronger for the larger sample. The effect of major changes on growth is slightly larger in absolute value than that of all government changes (in Table 4); this is true particularly in the smaller sample, where a change from zero to 1 in the probability of observing a major change of government reduces growth by 1.8 percent versus 1.4 in the case of G C HA N G E. In terms of standard deviations, however, the value of the coefficient of MJCHANGE is virtually identical to that of the coefficient of GCHANGE. These coefficients provide some weak evidence that there is actually a slightly larger effect of major government changes on growth than that of nonmajor changes. To see this, suppose that the nonmajor changes had a strong effect on growth. Since by moving from Table 4 to Table 5 we drop the nonmajor changes from the definition of change and include them in the no-change years, the coefficients on MJCHANGE in Table 5 should decrease somewhat from the coefficient on GCHANGE in Table 4. On the other hand, if nonmajor changes have no effect on growth, the coefficient on MJCHANGE should increase, since it is no longer diluted by the nonmajor changes. Since the government change coefficient does not decline in moving from Table 4 to Table 5, this suggests that nonmajor changes have less of a negative effect on growth than the major changes. As for the effect of growth on MJCHANGE, the three relevant coeff those on GROWTH, GROWTH(--1), and WGROWTH(--1)--all have the expected signs, though they are not statistically significant. On a priori grounds, we would have expected a significant negative effect of growth on the propensity of government changes. This somewhat inconclusive result should certainly not be taken as ironclad proof that growth has no significant effects on government changes, particularly given our different results for the

14 202 ALESINA, OZLER, ROUBINI, AND SWAGEL Table 5. Joint estimation of growth and major changes: base specification (t-statistics in parentheses) Sample Sample Major Major Growth Change Growth Change Equation Equation Equation Equation (1) (2) (3) (4) GROWTH (-1.54) (--0.45) MJCHANGE (-2.27) (- 1.86) CONSTANT (-1.61) (--9.86) (-1.77) (--9.01) GROWTH(-1) (0.18) (-0.24) (0.22) (-0.10) WGROWTH(-1) (5.99) (1.10) (6.29) (0.46) LATIN (- 1.96) (2.08) (- 1.22) (1.46) AFRICA (-0.19) (-0.19) (-0.58) (-0.20) EDUC (4.52) (1.74) EXADJ(-1) -- (0.057) (3.44) (2.64) MJCHANGE(-1) (0.96) (1.15) X 2 test of the 1 overidentifying restruction p-value Number of observations case of coups d'etat, discussed below. 22 Though not significant, the positive coefficient on W G R O W T H (-- 1) is consistent with the view that citizens may evaluate their government's performance with reference to the rest of the world. The coefficient on the Latin America dummy variable in the government change equation is positive and significant in the large sample. Recall from Table 1 that while the frequency of GCHANGE for Latin America is not higher than in the industrial and Asian countries, the frequency of MJCHANGE in Latin America is the highest in the world, almost double that of Asia and the industrial countries. It is also worth emphasizing that events concerning political unrest are likely to be under reported in African countries. Since Africa is a region with low growth and an underestimated measure of political instability, our result concerning the effect of political instability on growth would probably be strengthened by a correction of this under reporting bias. The chi-squared test of our overidentifying restriction does not reject our model of major government changes at reasonable levels of confidence, particularly in the large sample; the p-values are 0.66 in the large sample and 0.27 in the small sample. Table 6 presents our base specification for the case in which we take C O U P as the de-

15 POLITICAL INSTABILITY AND ECONOMIC GROWTH 203 Table 6. Joint estimation of growth and coups: base specification (t-statistics in parentheses) Sample Sample Growth Coup Growth Coup Equation Equation Equation Equation (1) (2) (3) (4) GROWTH (-6.27) (-5.18) COUP (-10.53) (-8.71) CONSTANT (-3.25) (-10.25) (-3.73) (-9.95) GROWTH(-1) (0.03) (0.04) (0.03) (0.04) WGROWTH(-1) (6.08) (4.89) (6.52) (4.47) LATIN (-1.37) (0.78) (-0.93) (1.37) AFRICA (-0.06) (-0.005) (-0.05) (-0.03) EDUC (1.98) (1.77) EXADJ(-1) -- (0.042) (1.95) (1.90) COUP(-1) (0.18) (0.20) X 2 test of the 1 overidentifying restriction p-value Number of observations pendent variable for government change. We continue to find a negative and statistically significant effect of the propensity of government changes (now coup d'etat) on growth. The t-statistics are larger and the coefficients smaller (in absolute value) than the corresponding values for MJCHANGE in Table 5. As discussed above, this implies that both M J CHANG E and C O UP have a significant effect on growth. In fact, in moving from Table 5 to Table 6 we include years of major changes (but not coups) in the no-change years, so that the average growth in no-change (that is, no-coup) years decreases. Thus the smaller coefficient on COUP in Table 6 does not necessarily indicate that coups have a smaller effect on growth than all major changes. Interestingly, with the C O U P definition of government changes, we now find that growth has a statistically significant effect on government changes, as shown by t-statistics on current growth, GROWTH, and lagged world growth, WGROWTH(-1). Moreover, the effect of growth on coups is substantial in magnitude. For the large sample, evaluating the explanatory variables at their mean values for Latin American countries (that is, GDP growth of 2.2 percent as shown in Table 1) and assuming that there was no coup the previous year, we find that the probability of a coup is only When growth is zero, however, the probability of a coup in the average Latin American country rises to 0.304, and when growth falls to -2 percent, the probability of a coup is If we assume the occurrence

16 204 ALESINA, 0ZLER, ROUBINI, AND SWAGEL of a coup the previous year, these probabilities rise to 0.098, 0.373, and 0.740, again with growth rates of 2.2 percent, 0.0 percent, and -2.0 percent, respectively. These results show that previous coups meaningfully increase the likelihood of a coup today but that this effect is swamped by the effects of growth. Simply put, countries with reasonable growth performance are unlikely to suffer coups regardless of other factors. While the results are most stark for Latin America, we find similar results for the effects of growth on coups in developing countries in other regions. Finally, an interesting observation emerges from the coefficients on the regional dummies LATIN and AFRICA in Tables 4, 5, and 6. While in Table 4 the coefficients on the two dummies are statistically significant with the negative sign typically found in the empirical growth literature, in Tables 5 and 6, these coefficients are insignificant. 23 These results suggest a plausible explanation for the unexplained low growth in these two regions. We find that controlling for the interaction of growth and political instability accounts for the relatively poor growth performance of Latin America and Africa. Our results differ substantially from those obtained by Londregan and Poole (1990) in their study of coups; unlike us, they find no effect of personal instability on growth. This difference in results most likely stems form differences in model specification. Our model differs from theirs in that we include data on world growth, education, and executive adjustments, providing a richer set of explanatory variables for both government changes and economic growth. Our test of our one overidentifying restriction shows that we do not reject our model for COUPS at significant levels of about 0.4 to 0.6. In contrast, Londregan and Poole (1990) are able not to reject their model at only the less-favorable 0.1 significance level. 7. Sensitivity Analysis and Discussion Our basic result that political instability is harmful to growth is robust to changes in the model specification. First, we investigate the effect of democratic institutions by adding our democracy variable, DEMOC, to our base specification of the growth equation (3). In results not shown, we find that after controlling for government changes, democracy is not a statistically significant determinant of growth--that is, democracy neither increases nor hinders economic growth. The lack of significance of the DEMOC variable holds in many other specifications. We find no significant changes in our results when we use additional lags of executive adjustments, EXADJ, or add the variable, ATTEMPT (or its lags), which measures unsuccessful attempts to change the government (such as aborted coups). This latter variable provides another proxy for the underlying signals of political instability that might increase the probability of a government collapse. We also add a dummy variable for industrialized countries, INDUST, to proxy for the high levels of human capital infrastructure in these nations. However, this variable is so highly correlated with the education variable EDUC that we can use only one or the other; since EDUC is a better proxy for the level of human capital, we rely on this variable in our basic specification. We also ran two specifications that included levels of GDP: one with the initial per capita level of GDP (from either 1950 or 1960 depending on the sample) and another with

17 POLITICAL INSTABILITY AND ECONOMIC GROWTH 205 the lagged level of GDP, GDP(-1). The effect of political instability on growth remains significant in all of these specifications. However, the fit of these models is not as good as that of the specifications in Tables 4, 5, and 6: the chi-squared statistics are generally much larger, indicating that the data reject the identifying restrictions. As an alternative means of identifying the model, we set the contemporaneous effect of growth on government changes to zero (note from Tables 4 and 5 that this parameter is not statistically significant). In two of these four cases, this specification results in a better fit (in terms of the test of the overidentifying restrictions), and a worse fit for the other two. In fact, the p-values are and for the large and small samples with GCH ANGE as the dependent variable, and and with MJCHANGE, which can be compared with the p-values of Tables 4 and 5. In all cases, however, we confirm our result that political instability has a statistically significant negative effect on growth. One interesting consequence of adding the restriction on the effect of growth on government changes is that the coefficient on lagged GDP growth becomes significant in both the "all changes" (GCHANGE) and "major changes" (MJCHANGE) specifications. This indicates that government changes are affected not only by current growth but also by a nation's growth performance in the recent past. Finally, we estimated our systems of equations leaving out oil-exporting countries and very small countries. Again, our results remain qualitatively unchanged. 8. Conclusions Political instability reduces growth. This result is particularly strong for the case of unconstitutional executive changes such as coups, as well for changes that significantly changes the ideological composition of the executive. The effect of instability on growth is less strong for the regular and frequent executive turnovers typical of industrial democracies. To some extent, and with some caveats, we also find that low growth increases the likelihood of government turnover, particularly in the case of coups d'etat. However, we cannot find any difference in the growth performance of democracies compared to nondemocracies. Finally the occurrence of a government change increases the likelihood of subsequent changes, suggesting that political instability tends to be persistent. We conclude by highlighting several possible extensions of this paper. First, it seems worthwhile to continue in our effort to distinguish cases of major government changes from routine turnovers of leadership that entail no significant changes in the ideological orientation of the government. This paper has been reasonably successful on this point, in the sense that the results using our new variable for major changes shed new light on the results obtained with the Jodice and Taylor (1983) variable. However, further research on this point is called for. For instance, one would like to address the following issue: in certain cases (Turkey in the late 1970s is a good example), frequent coalition reshuffling are indicators of an underlying political unrest (which, in the case of Turkey, culminated in a military coup in 1980). By disregarding such reshuffling, one underestimates the amount of political instability. In other cases (Italy in the post-world War II period is a good example), minor coalition reshuffling do not imply any significant amount of real political instability. Second, one might try to classify the ideological direction of government changes and test

18 206 ALESINA, OZLER, ROUBINI, AND SWAGEL for the effects of different governments' 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. Third, one may want to shed some more light on the channel linking political instability to growth. For instance, Alesina and Perotti (1996) suggest that the link goes through the effect of instability on investment. However, they use a different measure of instability. Exploring the linking channel and comparing our measure of political instability with others is a fruitful research objective. 9. Appendix A: Countries, Variables, and Sources Table A.1. Countries and sample periods Sample Sample Other a (60 Countries) (39 Countries) (14 Countries) United States Yugoslavia United Kingdom Haiti Austria Barbados Belgium Surinam Denmark Kuwait France Saudi Arabia Germany Syria Italy Hong Kong Luxembourg Nepal Netherlands Singapore Norway Algeria Sweden Botswana Switzerland Burundi Canada Cameroon Japan Cape Verde Finland Central African Republic Greece Chad Iceland Congo Ireland Gabon Portugal Gambia Spain Guinea-Bissau Turkey Ivory Coast Australia Lesotho New Zealand Liberia South Africa Madagascar Argentina Mali Bolivia Mauritania Brazil Mozambique Chile Niger Colombia Rwanda Costa Rica Senegal Dominican Republic Sierra Leone Ecuador Somalia El Salvador Swaziland Guatemala Tanzania Malta (1954) Jamaica (1953) Israel (1953) Jordan (1954) Bangladesh (1959) Indonesia (1962) South Korea (1953) Malaysia (1955) Benin (1959) Ghana (1955) Malawi (1954) Zimbabwe (1954) Sudan (1955) Zambia (1955)

19 POLITICAL INSTABILITY AND ECONOMIC GROWTH 207 Table A.1. (continued) Sample Sample Other a (60 Countries) (39 Countries) (14 Countries) Honduras Mexico Nicaragua Panama Paraguay Peru Uruguay Venezuela Guyana Trinidad Cyprus Egypt Sri Lanka Talwan India Pakistan Philippines Thailand Zalre Ethiopia Kenya Mauritius Morocco Nigeria Uganda Togo Tunisia Fiji Papua New Guinea Note: The end of the sample period is 1982 for all countries, except South Korea and Hong Kong, for which political data is available only through a. The date next to each country in this group indicates the beginning of the sample for which data are available. TableA.2. Variable definitions and sources Dependent variables GR 0 WTH Annual rate of growth of per capita GDP. Source: Summers and Heston (1991). GCHANGE MJCHAN GE COUP Government change. Dummy variable that takes a value of 1 for years in which there is either a coup or a regular government transfer and a value of zero otherwise. Source: Jodice and Taylor (1983), with several corrections of coding errors. Major government change. Dummy variable that takes a value of 1 for years in which there is either a coup or a major regular government transfer, and value of 0 otherwise. Source: authors' construction from Banks (various issues). Coups. Dummy variable that takes a value of 1 for years with at least one successful coup d'etat and zero otherwise. Source: Jodice and Taylor (1983).

20 208 ALESINA, OZLER, ROUBINI, AND SWAGEL Table A.2. (continued) Explanatory variables DEMOC EXADJ ATTEMPT GDP WGRO WTH EDUC LATIN AFRICA Democracy. Takes a value of 1 for democratic regimes, 2 for regimes mixing democratic and authoritarian features, and 3 for authoritarian regimes. Source: authors' construction from Banks (various issues) Executive adjustments. The number of changes in the composition of the executive not resulting in government transfers. Source: Jodice and Taylor (1983). Number of unsuccessful attempts at changing the government. Includes failed coups. Source: Jodice and Taylor (1983). Log of real per capita GDP, adjusted for terms of trade. Source: Summers and Heston (1991) World business cycle. GDP-weighted per capital GDP growth for the G-7 countries (United States, Japan, Germany, France, United Kingdom, Italy and Canada). Percentage of school age population enrolled in primary school. Source: World Bank Economic and Social Data Base. Missing years are obtained by piecewise linear interpolation. Dummy variable for countries in South American and Latin America. Dummy variable for countries in Africa. INDUST Dummy variable for industrial countries (IMF country code < 200). Acknowledgments We are grateful to Gary King and John Londregan for providing data and for useful conversations, to Robert Barro, Rudi Dornbusch, John Helliwell, Nancy Marion, Howard Rosenthal, Jennifer Widner, and numerous seminar participants for helpful comments, and to an anony- mous referee and an associate editor for many comments and suggestions. We thank Gina Raimondo and Jane Willis for excellent research assistance. For financial support, we thank IRIS, the Sloan Foundation, the UCLA Academic Senate, the Yale Social Science Research Fund, and the Hoover Institution. The usual disclaimer applies. The data and program used in this paper can be obtained on the Internet by anonymous FTP from NBER.Harvard.edu. Notes 1. The time period for our sample is limited by the availability of data on various indicators of political unrest; in particular, for almost haif of the countries, data are available starting only in See Appendix A for more details on the sample of countries. 2. Alesina and Rosenthal (1995) also address the issue of the joint determination of government changes and growth for the United States. 3. In the working paper version of this paper (Alesina et al., 1992) we provide a comparison with this literature by using our definition of political instability in cross-section growth regressions.

21 POLITICAL INSTABILITY AND ECONOMIC GROWTH Rodrik (1991) examines the effect of policy uncertainty on private investment in developing countries. Goodrich (1992) finds that foreign direct investment in less-developed countries is negatively affected by a high degree of political instability. 5. Kramer (1971), Fair (1978), Fiorina (1981), and Alesina and Rosenthal (1995) examine the United States, while Robertson (1983) and Lewis-Beck (1988) study other democracies. See Hibbs (1973) and Londregan and Poole (1990) for nondemocratic regimes. 6. This may occur, for instance, because the public cannot perfectly distinguish the negative effect of the exogenous shock from an inadequate policy response on the part of the government. Alesina, Londregan, and Rosenthal (1993) and Alesina and Rosenthal (1995) develop and then test this argument for the United States. 7. See Aizenman and Marion (1993) for a study of the effects of policy uncertainty on growth. 8. See Helliwell (1994) and the references cited therein. Barro (1996) finds strong effects from economic development to democratization but weak effects linking democratic institutions to growth. 9. We have corrected several mistakes in coding found in the publicly available version of this data set. 10. A set of criteria were used to distinguish a significant from an insignificant government change. The coding was done before any regressions were run and was never changed afterward. 11. An example of a country with many nonmajor government changes is Italy, with its sequence of governments formed by the same parties, sometimes even with the same prime minister. 12. An example of a country coded 2 is Mexico, where elections are regularly held but until recently there were widespread allegations of electoral fraud. In general, when judgment calls had to be made, we chose to be conservative in what we defined a democracy. 13. In contrasting these summary tables, we use only those observations for which data are available for the regression specifications we discuss below. Because our specification includes a lag of some variables, we drop forty-one observations for which data from 1959 is not available. This has only a trivial effect on the samples statistics. 14. F-tests reject the null hypothesis that these differences are not statistically significant. 15. The equation for government change is a simultaneous equations probit: Prob(c = 1) = Prob(c* > O) = 9 (ctcxc +flcx + YcY), where c is an indicator that equals 1 in years with a government change and zero otherwise, and 9 is the cumulative distribution function of the standard normal distribution. 16. An alternative way of identifying the system of equations is to impose the restriction Vc = 0 or yy = 0. See Section 7 for more on this point. 17. Own-country growth is excluded for the G-7 nations. 18. For the specifications with C O U P, we actually use the lagged number of coups rather than the dummy variable that remains the dependent variable; this provides for a slightly better fitting model (in the sense discussed below), though it does not noticeably change the result. The number of changes and major changes is not available, so for these we use the lagged dummy variable. 19. The difference in the level of significance of this coefficient in the two samples may be due to the fact that African countries are largely excluded from the large sample. In order to investigate this issue we estimated the system for the small sample excluding the African countries. The t-statistic of the coefficient of changes on growth becomes -2.4 and the fit generally improves. 20. These probabilities correspond to equation (2), where we use the estimated coefficients in the second column of Table 4 and the average values in the data for the corresponding variables, in this case the average values for the developed countries. 21. A high value for the test statistic (and correspondingly, a low p-value) would have indicated that data reject our identifying restrictions, casting doubt on our results. 22. Furthermore, an associate editor of this journal has suggested that our specification may tend to downplay the significance of the effect of growth on government changes. In fact the only variable that enters in the growth equation but not in the change equation is EDUC, a variable that changes slowly over time. Thus, EDUC may not be the ideal instrument to pick up year to year variations in growth, which might influence the propensity to government collapses. 23. The only exception is the borderline coefficient on the large sample for LATIN in Table 5.

22 210 ALESINA, t)zler, ROUBINI, AND SWAGEL References Aizenman, J., and N. Marion. (1993). "Policy Uncertainty, Persistence and Growth" Review of International Economics I, Alesina, A., J. Londregan, and H. Rosenthal. (1993). "A Model of the Political Economy of the United States?' American Political Science Review 87, Alesina, A., and R. Perotti. (1996). "Political Instability, Income distribution and Investment?' European Economic Review. Forthcoming. Alesina, A., and H. Rosenthal. (1995). Partisan Politics, Divided Government and the Economy. Cambridge: Cambridge University Press. Alesina, A., S. 0zler, N. Roubini, and P. Swagel. (1992). "Political Instability and Economic Growth." Working Paper 4173, National Bureau of Economic Research. Amemiya, T. (1978). "The Estimation of a Simultaneous Equation Generalized Probit Model?' Econometrica 46, Banks, A. (Various years). Political Handbook of the World. Various issues. Barro, R. (1991). "Economic Growth in a Cross Section of Countries?' Quarterly Journal of Economics 106, 407 A.A.A.. Barro, R. (1996). "Democracy and Growth." Journal of Economic Growth 1, Benhabib, J., and M. Spiegel. (1994). "The Role of Human Capital in Economic Development: Evidence From Aggregate Cross Country Data." Journal of Monetary Economics 34, Cukierman, A., S. Edwards, and G. Tabellini. (1992). "Seigniorage and Political Instability?' American Economic Review 82, Easterly, W., and S. Rebelo. (1993). "Fiscal Policy and Economic Growth: An Empirical Investigation?' Journal of Monetary Economics, 32. Edwards, S., and G. Tabellini. ( 1991 ). "Explaining Fiscal Policies and Inflation in Developing Countries." Journal of International Money and Finance 10, Fair, R. (1978). "The Effects of Economic Events on Votes for Presidents." Review of Economics and Statistics 60, Fiorina, M. (1981). Retrospective Voting in American National Elections. New Haven, CT: Yale University Press. Goodrich, S. (1992). "Political Instability as a Determinant of U.S. Foreign Direct Investment." Senior thesis, Harvard University. Grossman, H. ( 1991 ). "A General Equilibrium Theory of Insurrections,' American Economic Review 81, Gupta, D. (1990). The Economics of Political Violence. New York: Praeger. HelliweU, J. (1994). "Empirical Linkages Between Democracy and Growth." British Journal of Political Science 24, Hibbs, D. (1973). Mass Political Violence: A Cross-Sectional Analysis. New York: Wiley. Huntington, S. (1968). Political Order in Changing Societies. New Haven, CT: Yale University Press. Jodice, D., and D. L. Taylor. (1983). World Handbook of Social and Political Indicators. New Haven: Yale University Press. Kormendi, R. C., and P. G. Mcguire. (1985). "Macroeconomic Determinants of Growth: Cross-Country Evidence." Journal of Monetary Economics 16, Kramer, G. (1971). "Short Term Fluctuations in U.S. Voting Behavior, " American Political Science Review 65, Lewis-Beck, M. (1988). Economics and Elections. Ann Arbor: University of Michigan Press. Londregan, J., and K. Poole. (1990). "Poverty, the Coup Trap, and the Seizure of Executive Power." World Politics 42, Londregan, J., and K. Poole. (1991 a). "The Seizure of Executive Power and Economic Growth: Some Additional Evidence?' Unpublished paper. Londregan, J., and K. Poole. (1991 b). "Leadership Turnover and Unconstitutional Rule?' Unpublished paper. Mauro, P. (1993). "Country Risk and Growth: An Empirical Study Based Upon Subjective indices." Unpublished paper, Harvard University. Murphy, K., A. Shleifer, and Vishny. (1991). "The Allocation of Talent: Implications for Growth?' Quarterly Journal of Economics 106, Newey, W. K. (1987). "Efficient Estimation of Limited Dependent Variable Models with Endogenous Explanatory Variables." Journal of Econometrics 36,

23 POLITICAL INSTABILITY AND ECONOMIC GROWTH 211 Olson, M. (1982). The Rise and Decline of Nations: Economic Growth, Stagflation and Social Rigidities. New Haven: Yale University Press. Ozler, S., and G. Tabellini. (1991). "External Debt and Political Instability" Working paper, National Bureau of Economic Research. Robertson (1983). "Inflation, Unemployment and Government Collapse" Comparative Political Studies 15, Rodrik, D. (1991). "Policy Uncertainty and Private Investment in Developing Countries" JournalofDevelopment Economics 36, Roubini, N. (1991). "Economic and Political Determinants of Budget Deficits in Developing Countries" Journal of International Money and Finance, 10, Shleifer, A., and R. Vishny. (1993). "Corruption" Quarterly Journal of Economics 108, Summers, R., and A. Heston. (1991). "The Penn World Table (Mark 5): An Expanded Set of International Comparisons, " Review of lncome and Wealth, 34, Terrones, M. (1989). "Influence Activities and Economic Growth" Unpublished paper.