From Rags to Ri es: The Economics of Deprivation, Con ict and Welfare State

Transcription

1 From Rags to Ri es: The Economics of Deprivation, Con ict and Welfare State Dominic Rohner University of Cambridge February 8, 2006 Abstract Historical evidence has taught us that it is often the poorest and most deprived people in unequal societies who are recruited to ght in civil wars. The present contribution constructs a theoretical model of the choice between appropriation and production, where con ict is represented as mutual stealing. Fully speci ed production functions allow for both symmetrical outcomes and for introducing inequalities in abilities and endowments. It is shown that people with lower marginal returns to productive activities due to lower ability, less endowments or discrimination are more likely to choose appropriative activities. Further, it is examined theoretically under which conditions welfare state policies such as redistribution of income and capital, as well as education, health and poverty-alleviation spending can lead to less appropriative activities. Finally, the model s implications are tested empirically using logit estimations. JEL Classi cation: D02, D74, H50, I30. Keywords: Con ict, deprivation, welfare state, poverty, appropriative activities. Introduction History has shown us, that people lacking opportunities for escaping poverty are associated to civil wars, civil unrest, warlordism and other forms of political violence. For example, in countries such as South Africa, Rwanda, Guatemala, Nicaragua, El Salvador or Iran deprivation and inequality played a major role in the occurrence of political violence during the last century (cf. Muller and Seligson, 987, for a discussion). If the marginal gain from political violence 0 Address: Faculty of economics, Sidgwick Avenue, Cambridge CB3 9DD, United Kingdom. dr296@cam.ac.uk. Acknowledgements: I would like to thank Partha Dasgupta, Jun Xue and Steven J. Brams for their especially useful comments. As well helpful discussions with and comments of Toke Aidt, Mariya Aleksynska, Christoph Arnold, Ste an Ball, Paul Collier, Bruno S. Frey, Erik Gartzke, Anke Hoe er, Matthew Jackson, Pramila Krishnan, Simon Loertscher, Hamish Low, Henri B. Meier, Paul Milgrom, Hervé Moulin, Andreas Müller, Damjan Pfajfar, Alexander

2 (and appropriative activities in general) is higher than the marginal gain of joining the regular market economy, people have incentives to join warlords or guerilla movements. Of course, as well political and religious ideologies and the potential of organising rebellion matter in con icts, but grievances such as poverty and inequality make people more vulnerable to extremist doctrines and more easily recruitable by warlords. Thus, these di erent explanatory factors of con ict are not substitutes, but complements. The present contribution aims to explain, with the help of a game-theoretic model, the role played by deprivation in the occurrence of con ict, if property rights protection is imperfect. It will also be shown under what conditions welfare state institutions can lower the risk of appropriative con ict. This paper builds up on the growing theoretical literature in economics of con icts (cf. for example Schelling, 960; Skaperdas, 992; Hirshleifer, 995; Baker, 2003; Rohner, 2006). As well the literature on the welfare state is relevant for the present article. Most economic contributions have emphasised the impact of the welfare state on distorted incentives and on decreased competitiveness (cf. for example Agell, 996; Alesina and Perotti, 997; Lindbeck et al., 999). Further, the results of several empirical studies indicate that it makes sense to link, as it is done in my present contribution, the issues of con ict, deprivation, inequality and welfare state institutions: Inequality (Deininger, 2003), poverty (Collier and Hoe er, 2004; Fearon and Laitin, 2003) and lack of education (Deininger, 2003) have been found to increase the risk of con ict and crime. Justino (2005) has found that redistributive policies have been e ective in reducing unrest on the local level in India. Despite these empirical results, only relatively few theoretical models focussing on those issues have been built. The contributions of Grossman (995), Azam (200) and Noh (2002) examine the deterrent e ect of income transfers on appropriative activities. Brito and Intriligator (985) link the possibility of avoiding inter-state wars through resource transfers with the issue of imperfect information, and Grossman (994) puts emphasis on the fact that land reforms can result in less appropriative activities. However, most of these interesting contributions use a Grossman-style model in which the inequality between the two types of players is inherent in the model. As the roles, for example, predator versus prey, or landowner versus Plekhanov, Emiliano Santoro, Carlos Seiglie, Kevin Sheedy, Diego Winkelried and of the participants of the "Shadow Talk" at the University of Cambridge in 2005, of the "Jan Tinbergen Peace Science Conference" in Amsterdam in 2005 and of the "Workshop on Game Theory" in 2005 in Bolzano are gratefully acknowledged. I would also like to thank Havard Strand, from PRIO (International Peace Research Institute Oslo) who has kindly agreed to share updated PRIO-con ict data with me. As de ned in the present contribution, appropriative activities include all actions taken in the aim of altering the existing distribution of social wealth in favour of the player who takes the action. Examples of appropriative activities include ghting in civil wars, joining warlord movements, committing political violence targeted to alter the existing wealth distribution, criminal activities, rent seeking, lobbying. 2

3 peasant families, are assigned from the beginning, these models cannot account for a symmetric outcome, with all players choosing some part of appropriative activities. Moreover, these models focus on one policy issue at the time and do not allow for considering the impact of various welfare state institutions within one single model. Furthermore, these contributions focus mainly on income transfers, but do not treat other welfare state policies such as capital redistribution 2, education 3, health policy and anti-poverty-programmes. Thus, there is still an important need for further research linking the issues of con ict and welfare state. The contribution of my present paper is of three sorts: First, I will build a con ict model which will permit for symmetrical outcomes and which will include a fully speci ed production function, enabling us to consider the impact of several distinct welfare state policies at the same time. Second, I will not only focus on income taxation, but I will as well analyse the cases of capital redistribution, education spending, health policy and antipoverty-programmes, which have as far as I know never been discussed in the context of our research question. Third, the impact of di erent welfare state policies will be assessed empirically with the help of pooled panel data for 208 countries and 44 years. Up to now only case studies for particular countries exist in the literature. The remaining part of the present contribution is organised as follows: Section 2 builds a simple model of appropriative con ict, section 3 focusses on the impact of di erent welfare state policies, and section 4 discusses some extensions to the basic model. Section 5 is devoted to an empirical test of the e ects of welfare state policies on con ict and section 6 concludes. 2 The model As for armed con icts usually large groups of the population of a given country or region are involved, a 2-player framework seems appropriate. Each player represents the aggregate choice of individuals being part of two larger groups of the society. For example, one player could represent the black people, and the other player the white people. Alternatively, one player could correspond to educated people with a high ability, and the other player to illiterate people from poor rural areas. In the aim of keeping the model analytically tractable, collective action issues are ignored in the present framework. The fact of not treating the decision process of the leader explicitly is not very problematic. It is reasonable to think of the level of appropriation chosen by the government as a weighted average of the optimal level of appropriation for the population and of the optimal level 2 The model studying land reforms in Grossman (994) is quite speci cally designed for treating the factor of production "land". It cannot easily be generalised for other kinds of endowments, such as capital. Similarly, Brito and Intriligator s (985) model focusses on the question of imperfect information, and is not designed for analysing the impact of welfare state policies, such as capital redistribution. 3 Grossman and Kim (2003) link the issues of education and con ict. However, they focus on explaining di erences across countries in educational policies. 3

4 for the ruling elite. The government needs both the support of the population and of the elite to stay in o ce and the relative weights given to the population and the elite depend on how democratic a country is. But in any case, policy shocks which decrease the optimal level of appropriation for the population also decrease in such a framework the level of appropriation chosen by the government (although obviously to a smaller extent). We consider a population of two players having the choice between productive (L) and appropriative (F) activities. As de ned on page (footnote), appropriative activities are unproductive, aim to alter the existing distribution of social welfare, and constitute a zero-sum-game. Thus, even though they may be utility maximising for a particular player, they are socially harmful and constitute a net deadweight loss on the level of the society as a whole. There is only one output good, y, which is produced using two inputs, labour (L) and capital (K). Whereas L is the choice variable, K is assumed to be a given endowment of capital. Typically, the payo of each agent consists of the non-stolen part of her own production plus the part of the opponent s production which she steals. For simplicity, the appropriated part of the other player s production depends linearly on the time share a player spends for appropriative activities. I include a parameter for player (and for player 2) which represents a friction in the appropriation activity. The idea here is similar to the one of iceberg trade cost. A part of the stolen production from the other player is lost due to ghting or transportation to the predator country. We have the following payo (utility) functions for the two players: V i = ( F j )y i + F i y j = L jy i + ( L i )y j () subject to the constraint L i + F i =. Moreover, we can include Cobb-Douglas production functions y = L a K b and y 2 = L c 2K d 2, where ; =total factor productivities; a,b,c,d=parameters. The functions show decreasing marginal returns for the choice variable L, i.e. a<,b<,c<,d<. We get the following payo functions: V = L 2 L a K b + ( L ) L c 2 Kd 2 (2a) V 2 = L L c 2K d 2 + ( L 2 ) L a K b i equal to zero, we get the rst order conditions (the second order conditions hold). Reformulating, the following expression is obtained for player (it is similar for player 2): L = a b K K d 2 L c 2 (3) Taking into account the constraint of L i, we can re-write the reaction function (3) in the following way: 4

5 8 >< a K b L = K d 2 L c 2 if >: if a K b a K b K d 2 L c 2 K d 2 L c 2 > (3 ) We can easily see that the higher the total factor productivity and the bigger the capital endowment of a player are, the bigger proportion of her time will be spent for productive activities. This corresponds to the argument mentioned earlier, that in unequal societies the players with a lower total factor productivity or a lower capital endowment will be induced to spend a substantial fraction of her time for appropriative activities. Further, we can also see that if poor countries get richer on the whole (parallel increase in and ), the level of productive activities increases, and accordingly appropriation decreases. One could argue that this result does not correspond to what is observed empirically, as often rebel leaders, suicide bombers and "star" terrorists like Osama Bin Laden are well-educated and come from relatively wealthy families. However, the present model is about clashes between di erent subgroups of a country, and not about individual decisions of leaders. On the aggregate level the conclusion that rebel movements start generally in disadvantaged groups holds empirically. The fact that the rebel leaders and suicide bombers sometimes steam from well-o families can be explained by the huge supply of people in disadvantaged regions being willing to join the rebel movement. Among this big number of candidates, the movement chooses the best-educated and most skilled ones which are likely to be of a relatively privileged family background. For describing graphically the Nash equilibria, we can, without loss of generality, rst focus on a symmetrical case where a = c; b = d; = ; = : Figure 4 displays the reaction functions for particular values of the di erent parameters. We will rst consider the case of a ghting-trap. For a j < (as in the case of the dotted curves), there is only one single Nash equilibrium, (0,0). If player s (black) reaction function rotates towards north-west, and player 2 s (grey) reaction function rotates towards south-east, we will eventually end up in the case where a K b i K d j =, with the two Nash equilibria (0,0) and (,). This corresponds to the solid lines. The outcome (,) is Pareto-superior to (0,0). Continuing to rotate the reaction functions further (not displayed in gure ) will lead to a case, where a K b i K d j >. In this situation we obtain three Nash equilibria, (0,0), (,), and an intermediate one. Figure 2 describes the basins of attraction for the case with three equilibria. The two extreme value equilibria E (0,0) and E (,) in gure 2 are stable, 9 >= >; K b i K d 4 Case : a=0.5,b=0.5, = 0:5; = 2; K = ; K 2 =. The thin black dotted line is the reaction function of player, whereas the thin grey dotted line refers to the reaction function of player 2. Case 2: a=0.5, b=0.5, = 0:5; = 4; K = ; K 2 =. The thick black solid line represents the reaction function of player, the thick grey solid line corresponds to the reaction function of player 2. 5

6 L L2 Figure : Reaction functions for di erent parameter values L E'' = (,) E''' RF2 RF E' = (0,0) L2 Figure 2: Basins of attraction for the case of three equilibria 6

7 whereas the intermediate equilibrium E is unstable. Towards the south-west of the intermediate equilibrium is the basin of attraction of the "bad" equilibrium E (0,0) and in the north-east of the intermediate equilibrium is the basin of attraction of the "good" equilibrium E (,). The more RF rotates to the north-west and the more RF2 rotates to the south-east, the more the intermediate equilibrium E moves to the south-west. This leads to the basin of attraction of the "bad" equilibrium becoming smaller and to the basin of attraction of the "good" equilibrium becoming larger. The focus in the next section will be on how to achieve the "good" outcome (,), if initially the country is stuck in the " ghting-trap" (0,0). Therefore, it is assumed that initially a K b i K d j <. It will be shown that under certain conditions welfare state policies can be able to rotate player s reaction function towards north-west, and to rotate player 2 s reaction function towards south-east in gure. If the policy shock is big enough, it might be possible to achieve multiple equilibria, whereas (0,0) is still possible, but (,) is now a potential outcome as well. Thus, the framework becomes a "coordination"- game. Further, as seen before, the more RF rotates to the north-west and RF2 rotates to the south-east, the smaller becomes the basin of attraction of the "bad" equilibrium and the larger becomes the basin of attraction of the "good" equilibrium. In this case the likelihood of ending up in a " ghting-trap" decreases, and it becomes more likely to achieve the good outcome (,). As explained above, rotations of RF to the north-west and of RF2 to the south-east are always a good thing in terms of political stability. Therefore, for the comparative statics section it is most convenient to focus on the rotations. However, for the sake of completeness, we can compute the exact analytical value of the Nash equilibria, by plugging one reaction function in the other. The point (0,0) is always a possible equilibrium. The intermediate equilibrium value for L is given by equation (4) (it is analogous for player 2). L = " a c ()( c) c ( K b b( c) ()( c) ) K2 d( ) ()( c) # ()( c) ()( c) (4) The present analysis is based on the framework of a static game. If the game was repeated, the chances of ending up in the "good" equilibrium could be increased through strategies such as "grim strategy" or "tit-for-tat". 3 Comparative statics of welfare state policies In this part we will focus on the impact of di erent welfare state policies on the players decisions between productive and appropriative activities. We should note that the rst-best policy would be the enforcement of property rights, eliminating appropriative activities altogether. Accordingly, welfare state institutions represent only a second-best policy that becomes important when 7

8 property rights protection is imperfect. We can regard the society as being composed by two types of agents: One "high productivity" type and one "low productivity" type. Several reasons could be invoked for "low productivity" players getting a lower payo per invested amount of labour time: First, the total factor productivity (respective ) could be lower due to natural (genetic) abilities, such as intelligence, or due to discrimination against ethnical or religious minorities (i.e. to bigger di culties of getting an appropriate job). Similarly, the family background (wealth, class) could be the reason for the di erent productivities. Children from rich families could get a better education in private schools rather than in state schools, which would help them develop better abilities for producing. Second, the lower return to labour for the "low productivity" types could be due to smaller capital endowments. Being from a rich family would result in a bigger availability of capital and as capital is included in the production function, there is a higher marginal productivity of labour for a higher amount of capital (as stated in standard neoclassical economic theory). In the framework of our model, "low productivity" types would typically have higher incentives than "high productivity" ones for choosing appropriative activities. Welfare state policies could prevent "low productivity" types from ghting. 3. Income taxation for transfers First, we will focus on transfers from the high-income player to the low-income player by the mean of income taxation. In this case, the payo function of player becomes (it is analogous for player 2): V = L 2 ( t)y + ( L ) [( t)y 2 ] + t(y + y 2 ) 2 Introducing the production function in (5), we obtain: (5) V = L 2 ( t)l a K b + ( L )( t) L c 2 Kd 2 + t(la K b + L c 2K d 2 ) 2 (5 ) From the rst order conditions follows reaction function (6) for player (for player 2 the result is similar): " ak b L = L c 2 Kd 2 # ( t)l2 + t 2 ( t) Taking the rst derivative of (6) with respect to t, we can see under which conditions more income taxation leads to more or less appropriative activities. As we are interested in the direction of the shift of the reaction function, we have to focus on the change in L for a change in t, holding L 2 constant at a given level. The approach will be similar in the other subsections. (6) 8

9 # " = ( ) ak b ( t)l2 + t 2 L c 2 Kd ( t) 2 " 2 ( )L 2 + t # 2( t) ( t) a (7) If the expression (7) is positive or negative depends exclusively on the last term in (7) (as all other terms are unambiguously positive). More precisely, the numerator of the last term in (7) is decisive. Thus, we get the following proposition: Proposition A marginal increase in income taxation has a positive impact on the producing time of player (rotates his reaction function towards northwest in gure ) if 2 ( )L 2 + t 2( t) > 0, i.e. for a low L 2, for a high and a high t. For a high, which corresponds to a low friction of stealing, more redistribution decreases the incentives for appropriative activities. Intuitively, if a player knows that she will get a substantial part of the other player s production through redistribution, she will have less incentives to steal her own future income. For a very low, stealing becomes in any case less of an issue, and the dominant impact of a rise in taxes are the reduced incentives for productive activities. The situation is similar for the level of L 2. If we are in a ghting-trap (low level of L 2 ) the disincentives of stealing ones own future tax revenue dominate, whereas if a country is doing well (high level of L 2 ), the disincentives of taxation on production e ort dominate. This corresponds to the situation in most industrial nations nowadays. One can imagine the redistribution process as institutionalised "stealing" from the rich for giving it to the poor, managed by the state. Unlike redistribution through appropriative activities, redistribution through the state does not involve a friction cost of ghting and does not lead to an opportunity cost of time spent for appropriation. For these reasons, income redistribution through a welfare state is Pareto-superior to income redistribution through appropriation. It is important to note that we have not taken the increase in the demand for "leisure" and the administrative costs caused by redistribution into account. These factors can result in a deadweight loss. Moreover, in a n-player rather than a 2-player framework stealing from the opponent would only to a smaller extent lower the aggregate taxation revenue. Therefore, the incentives of renouncing to steal would be less important for a given player. An interesting variation of the policy of forfeit income transfers would be to make the transfers dependent on the receiving player choosing a zero (or very low) level of appropriation. Such conditional transfers could enhance the incentives of behaving in a peaceful way. 9

10 3.2 Redistribution of capital Each players time spent for productive activities increases for an increase in his own capital and decreases for an increase in the opponent s capital. We can easily see this by taking for equation (3) the rst derivative of L with respect to K (displayed in equation (8)) and K 2 (displayed in equation (9)). It is similar for the reaction function of the second = 2 b d a a+b K a b K K d 2 L c K a d 2 L c 2 > 0 (8) 2 < 0 (9) Thus, the impact of capital redistribution in the present framework is ambiguous 5. If capital is redistributed from player to player 2, typically player 2 has smaller incentives for appropriative activities, but player s incentives for ghting increase. Simple capital transfers between large population segments do not appear to be an e cient policy for avoiding con ict. This conclusion seems reasonable. History has taught us that it might not always be a good policy to redistribute capital which is productively employed. Even though equal land distribution is de nitively a concern, Zimbabwe s expropriation strategy, for example, has reduced economic growth, but has not contributed to more political stability. The predictions of the present model with respect to capital distribution contrast with the result of Grossman (994) who concludes that redistribution of land can decrease con ict. The di erent policy conclusions of the two models are due to the fact that the Grossman-model is asymmetric and attributes the role of the prey to the landlords and the role of the predator to the peasants. Typically, in such a framework redistributing land from the very few landlords to the huge masses of peasants will reduce the con ict potential of the latter without increasing the risk of rebellion of the former. This seems sensible, and the Grossman-model is surely very appealing for analysing situations of extreme wealth disparities among classes. However, our model focusses rather on cases in which two big groups of the society face each other, such as the white and the black population in South 5 The impact of capital redistribution can be assessed in a more formal way. If the intermediate equilibrium moves to the south-west (which corresponds to a decrease of L nd L 2 ), the basin of attraction of the good equilibrium increases and the basin of attraction of the bad equilibrium decreases. This leads to a lower risk of con ict. Formally this corresponds to the following condition for player. L 0 " = # c a c ()( c) (K + t) ( b b( c) ()( c) ) d( ) ()( c) (K 2 t) ()( c) ()( c) " # c <L = a c ()( c) K ( b b( c) ()( c) ) d( ) ()( c) ()( c) ()( c) K2. For the second player it is similar. 0

11 Africa or the Hutu and the Tutsi in Rwanda. In such contexts, typically, taking away capital or land of one group can lead to increased aggression of this part of the population. In Rwanda for example, the Hutu majority usurping the Tutsi minority at the beginning of the 990s led to the organisation of the Tutsi resistance under the Rwandan Patriotic Front (RPF). This illustrates that when two big groups of the population are involved, the capital and land transfers of one group to another are likely to lead to armed resistance of the group which looses from the transfer. The situation is di erent for the redistribution of capital which is not employed in production, but which is consumed. We can assume for now that there is a maximum amount of capital which the production technology allows for. Thus K i = K i for K i < K, and K i = K for K i K. It follows that if, say, K 2 K, (marginally) redistributing capital from player 2 to player increases player s productive capital without reducing player 2 s productive capital. In this case, redistribution from player 2 to player can lead to less appropriative activities. In this second framework, redistribution of capital could take the form of a tax on certain luxury goods. The ndings of this subsection are summarised in the following proposition: Proposition 2 The impact of capital redistribution is ambiguous in the general framework. For the special case of capital which is consumed rather than invested, redistribution can lead to less appropriation. 3.3 Education and public health promoting and povertyalleviation programmes In this part it is examined what happens if taxation revenue is not simply transferred, but spent for education, health and poverty-alleviation programmes, which would eventually lead to a higher total factor productivity. Thus, we have the following payo function for player (for player 2 it is analogous): V = L 2 ( t)y + ( L ) [( t)y 2 ] (0) Introducing the production function in (0), we get: V = L 2 ( t)(t)l a K b + ( L )( t) ((t)) L c 2 Kd 2 (0 ) Please note that now and are strictly increasing functions of t, i.e. = (t) and = (t). We obtain reaction function () for player (it is similar for player 2): L = ((t)) ((t)) a b K K d 2 L c 2 ( t) () The partial derivative of L with respect to t is positive under the > t : This means that education, health and anti-poverty spending are likely to increase the labour time of a given player if she has a low initial level of total

12 factor productivity and if the policy measures increase her productivity by much. By contrast, the increase in the total factor productivity of the opponent tends to decrease the rst player s incentives to pursue productive activities. This is intuitive, as an increase in the productivity of the opponent makes the opponent a more attractive target. However, for most parameter values it is likely that these three policies lead to less con ict overall, as the con is multiplied by, whereas the con ict-enhancing term is multiplied by <. This reasoning becomes more apparent, if we focus on the symmetrical case, where =. In this case, the condition for education, health @ spending to reduce con ict > (t) ( t). It follows from our assumptions that taxation will have a positive impact on the total factor 0. It is furthermore reasonable to assume that the marginal productivity of taxation spending is strictly Thus, 2 (@t) 2 < 0. For low levels of and t is big. Thus, for low values of and of t it is likely > (t) ( t) holds. The policy implications for reducing con ict, i.e. to increase spending for education, health and povertyalleviation programmes if total factor productivity is low, appear reasonable. Proposition 3 summarises our reasoning. Proposition 3 When the total factor productivity is a strictly increasing function of education, health and poverty-alleviation spending, marginally increasing the public expenditures for these policies increases the overall time spent for productive activities if the country is "stuck" in a low productivity- and low taxation- equilibrium. Increases in education, health and anti-poverty spending bene t above all the "low productivity" player. This is the case because the marginal productivity of those policies is decreasing. The "low productivity" player is in general the most likely to choose a high level of appropriative activities. Therefore, education, health promotion and poverty-alleviation programmes are well-focused. Furthermore, independently of the impact of an increase in the total factor productivity on the labour supply of the two types, such an increase in the total factor productivity could result in a higher output y i. In fact, there will be a trade-o between the costs of these kinds of public expenditures and their returns. However, this issue is already well-documented in the literature and is not directly linked to our main focus. 2

13 4 Con ict analysis in a two-period framework and the impact of education and of equality of chances In this section the basic model will be extended to a two-period framework. The basic idea is that education spending at an early stage and the o ering of perspectives for a bright future can prevent the disadvantaged of investing in ghting capital. In the rst period the player lives in a poor rural setting. Modest surplus from agricultural activity can be invested in either productive capital (in particular production skills and human capital) or in ghting capital. To put it di erently, the player only needs to spend a proportion s of her time for assuring her subsistence expenditures and has some spare time (-s) (the surplus) which can be used for acquiring productive or ghting skills. The choice in period one will in uence the total factor productivity and the ghting technology in period two. Equation (2) displays the choice of player in the rst period (it is similar for player 2). The time endowment is normalised to. ( s) = E + W (2) where s=part of income from time endowment in the rst period used for subsistence expenditures, E=accumulation of productive (human) capital, W=accumulation of ghting capital. This equation can be reformulated and be expressed in terms of E respectively W. E E = ( E + W s) = e( s) (3) W W = ( E + W s) = w( s) (4) where e=proportion of surplus which is used for the accumulation of productive capital, w=proportion of surplus that is used for the accumulation of ghting capital. The accumulation of productive capital in period one increases the total factor productivity in period two: = ( + E). Similarly, accumulating ghting capital in the rst period improves the ghting technology in the second period. In particular, it reduces the friction from appropriation. We have: = ( + W ). The framework in period two is identical to the one used in the one-period model of the previous section. Thus, in period two we have the same reaction function for player one as before in equation (3), with the di erence that and are now endogenous. The new reaction function is displayed in equation 5. 3

14 0 L ( E + )a A ( E + 2 ) ( W + ) K b K ( W + )d 2 L c( + W ) 2 (5) Player s production time in period two is increased by her own accumulation of productive capital in period one (E ), and is negatively a ected by the accumulation of skills of her opponent (E 2 ), as is in the nominator and enters the denominator. The intuitive reason is that increases in her own total factor productivity increase the incentives to work, whereas a higher total factor productivity of the opponent increases the incentives to steal from the opponent. However, as is in the power of, and is in the power of <, on the whole the positive impact of skills accumulation of both players outweighs its negative impact. For a symmetrical case both players will end up ghting less. In addition, the accumulation of human capital will as well lead to a higher economic output. As shown in equation 6, the rst derivative of the production time in period two with respect to the ghting technology parameter is negative. Thus, the greater the term (which corresponds to a lower friction of appropriation), the more time will be spent for appropriation in the second period. Accordingly, W, the accumulation of ghting technology in period one, increases the intensity of con ict in = ( h i ) a KK b d 2 L c 2 (ln + c ln L 2 + d ln K 2 ) (6) It follows that the level of con ict is lower the higher E is and the lower W is. Thus, a policy would be desirable if it increases the proportion of the surplus in the rst period used for human capital accumulation and accordingly reduces the ghting capital accumulation. Education and the o ering of a perspective for a better future could ful ll this role. If children spend more time in school, they acquire production skills which give them incentives later on to spend their time producing rather than appropriating. Further, the time they spend in school cannot be used for acquiring ghting capital. The issue of child soldiers in Africa illustrates well this problem. Lack of real chances in life and of a meritocratic society gives incentives to children to accumulate ghting/appropriation capital in their youth (period ). Later on they have incentives to use the ghting skills for appropriation. A possible policy approach is to tax an amount r of the surplus in period one and to use it for education spending. It would also be conceivable to administrate a direct or indirect bene t scheme for families who put their children in school. Taxing the amount r would leave a disposable income of ( s) r. The new values of E and W are displayed in the equations (7) and (8). E 0 = e [( s) r] + r = E + ( e)r > E (7) 4

15 W 0 = w [( s) r] = W wr < W (8) Thus, also in a two-period framework education spending reduces the risk of con ict. One can think of the present model with overlapping generations. At a given moment in time the parents bring up their children and decide on which activities to spend their children s time. The parents decision for the allocation of their own time is the one of period 2 in the present model, whereas the (simultaneous) decision of the allocation of their children s time is the decision of period. In the following period, the parents die and the children become themselves parents and so on. It seems reasonable to assume that parents who spend a large amount of time for appropriation are more likely to encourage their children to accumulate ghting capital rather than production capital. Further, children who have acquired ghting technology will become ghters and will provide a bad in uence to their own children. We are in the presence of a vicious circle. The case of the Italian ma a illustrates well this reasoning. Children of ma osi tend to join their parents criminal organisation. In this way, the ma a dynasty can survive for numerous decades. Enhanced education for young children seems an appropriate policy measure for breaking such vicious cycles, as it attacks directly the roots of the problem. 5 Empirical evidence Clearly, the main contribution of the present paper is to economic theory. However, it makes sense to test in an empirical section whether our theoretical ndings are reasonable. In order to keep the paper of a convenient length and for enhancing conciseness, this empirical part will be kept very short. In the literature there have been several in uential empirical studies of civil wars (see Fearon and Laitin, 2003; Collier and Hoe er, 2004; Miguel, Satyanath and Sergenti, 2004). These studies have focussed on socioeconomic explanatory factors of civil wars such as economic growth, wealth, ethnicity, natural resources, and population. Institutional and policy-related factors such as welfare state policies have received only very little attention in the empirical literature. By testing our theoretical model I would like to address this shortcoming. In this empirical part pooled panel data for 208 countries from 96 to 2004 will be used to explain the likelihood of outbreak of a civil war. As most papers in the eld, I will attempt to explain the outbreak of wars rather than the number of periods of ghting. As dependent variable, civil war data from the "Data on Armed Con icts" database of PRIO (2006, in collaboration with the University of Uppsala) is used. All internal and internationalised internal con icts that cause more than 000 battle related deaths for the whole con ict are coded as civil wars. This is a dummy variable which takes the value of for an outbreak of war, the value of 0 for peace and is coded as missing for ongoing wars. This way of proceeding is frequently used in the empirical literature on con icts. 5

16 The main explanatory and control variables are taken from various sources (explained in more detail in the appendix). The number of observations di ers for all included welfare state variables. Data on government spending is available for , on education spending for , on health spending for and on taxation revenue for An important issue is the direction of causality between outbreaks of civil wars on one hand and economic output and welfare state policies on the other hand. A negative correlation between these variables could be due to welfare spending and economic performance decreasing the risk of civil wars or alternatively due to the depressing e ect of war on the economy and on public spending. Miguel, Satyanath and Sergenti (2004) have solved this endogeneity problem with the help of instrumental variables (rainfall) for economic shocks. However, as in our case several variables could potentially show an endogeneity bias, this does not seem the ideal approach for our study. Collier and Hoe er (2004) have solved the endogeneity problem by estimating war outbreaks in ve year periods with the help of the values of explanatory variables at the beginning of those periods. This approach is not ideal in our case as it would lead to too small a number of observations for certain explanatory variables, e.g. health spending and taxation revenue. The chosen approach in the present paper is to use logit estimations for annual data and to use the rst lag of variables a ected by a potential endogeneity bias, i.e. the welfare state variables and the "economic" control variables. The results of the logit estimations of the impact of di erent welfare state policies on the risk of civil war outbreaks are displayed in table. In order to avoid multicollinearity, too strongly correlated variables or combinations of variables have not been included in the same speci cations of regressions. We can see that the con ict-reducing impact of government spending is signi cant at a % level for both the speci cations () and (2). Education is also found to decrease the risk of civil war outbreaks. It is signi cant at 5% in speci cation (3) and at 0% in speci cation (4). The variable of health spending in equation (5) has also a negative sign, but is not statistically significant. Finally, higher revenues of taxation also reduce the war risk at a level of con dence of 95%. All control variables have the expected sign, namely economic output and growth as well as civil liberties (small numbers indicate a high level of liberties) decreasing the risk of civil war, and a higher population and more ethnic fractionalization increasing the war risk. Several robustness checks have been performed. First of all, numerous di erent speci cations and several other control variables have been used, such as for example exportations of natural resources, religious and linguistic fractionalisation, several topographic and demographic variables and institutional factors. They did not a ect our ndings of a statistically signi cant con ict-reducing impact of government spending, education expenditures and of tax revenue. Our results have also been found robust to the inclusion of natural logarithms for GDP per capita and population, as well as for the inclusion of robust standard errors. 6

17 () (2) (3) (4) (5) (6) Governm. exp.( ) (3.6)*** (3.6)*** Educat. exp. ( ) (2.39)** (.72)* Health exp. ( ) (0.) Tax revenue ( ) 0.4 (2.6)** GDP per cap. ( ) (2.20)** (.8)* (0.82) GDP growth ( ) (2.09)** (.27) (.6) (.6) (0.48) Population (0.80) (.08) (.83)* (.9) (.55) (.47) Ethnic fract (.86)* (.33) (0.27) (.6) (2.89)*** Civil liberties (2.88)*** Constant (6.79)*** (6.20)*** (4.86)*** (3.89)*** (2.40)** (2.95)*** Observations Pseudo R Log Likelihood Absolute value of z statistics in parentheses: * significant at 0%; ** significant at 5%; *** significant at % Table : Logit estimations of outbreaks of civil wars For further robustness checks, as displayed in the columns () and (2) of table 2, updated con ict data from Collier and Hoe er (2004) and from Gleditsch (2004) have been used as dependent variable. The same control variables have been included as in the second speci cation of table. As well for this alternative civil war data the war-reducing impact of government expenditures remains signi cant at %. Also the implications of education spending and tax revenue remain valid. For updated Collier and Hoe er (2004) data, also health expenditures have a war-reducing impact which is signi cant at a 5% level. In column (3) logit estimates with corrected coe cients for rare events (using the method of King and Zeng, 200) are included. Also in this case government expenditures remain signi cant at a % level of con dence. Also the variables of education spending and of tax revenue remain signi cant when this corrected logit estimation is used. In column (4) I take 5 year periods as done in Collier and Hoe er (2004). Their updated civil war data is used. Also for this modi cation the impact of government expenditure remains statistically signi cant at a 0% level. Finally, instead of using lags, the possible endogeneity issue can also be addressed with the help of instruments. Column (5) displays the results for a two-stage probit estimation of the impact of government expenditures. The instruments used for that explanatory variable are the level of foreign aid, the altitude range (as mountain territory makes public service provision more di - 7

18 () (2) (3) (4) (5) Gleditsch Collier/ Rare 5 year Two stage data Hoeffler events periods probit data logit Governm. exp (3.36)*** (3.02)*** (2.83)*** (.72)* (2.54)** GDP per cap (2.02)** GDP growth (3.77)*** (2.40)** (.32) (.49) (3.4)*** Pop (.50) (.34) (.78)* (.83)* (0.45) Ethnic fract (.2) (0.89) (.95)* (.82) (.0) Civil libert (3.43)*** (3.32)*** (3.46)*** (.94)* Constant (7.8)*** (6.8)*** (5.80)*** (4.77)*** (0.42) Observations PseudoR Log Likelih Absolute value of z statistics in parentheses: * significant at 0%; ** significant at 5%; *** significant at %. For the variable "Government expenditures" in columns (), (2) and (3) its first lag is used, in column (4) its level is used, in (5) it is instrumented. For "GDP per capita" the level is used in (5). For "GDP growth", columns (), (2) and (3) refer to its first lag, (4) and (5) to its level. Table 2: Robustness checks for outbreaks of civil wars cult) and the proportion of people of 65 years and above. Government expenditures remain signi cant at a 5% level. Also the other Welfare state variables are robust with respect to the use of two-stage probit estimations. 6 Conclusion The present contribution has focussed on the impact of deprivation in unequal societies on the decision taking of players between appropriative and productive activities, and on the question of whether welfare state policies may be able to reduce the equilibrium level of appropriation. Con ict has been represented as mutual stealing, whereas a friction for stealing has been included. Using fully speci ed production functions has permitted for both symmetrical outcomes and for the inclusion of inequality in capital endowments and in total factor productivity. It has been shown that population groups with a lower total factor productivity spend more time for appropriation. Income taxation reduces the incentives for appropriation above all if the economy is stuck in a low-production trap. For higher levels of production, under certain conditions the undesirable impact of distortions and disincentives for productive work due to taxation may prevail. The impact of capital redistribution is ambiguous, decreasing the incentives to ght of the player who receives 8

19 capital, but increasing the incentives to ght for his opponent. However, redistribution of capital that is consumed rather than invested can lead to reduced ghting. Education, health and anti-poverty spending have above all an appropriation reducing impact if initially the total factor productivity is low. In a two-period framework education and perspectives for a better future result in children accumulating less ghting capital and more production skills, which leads to a lower level of con ict in the long-run. Further, the model has been empirically tested using logit estimations. Higher government expenditures, education spending and tax revenues have been found to lower the risk of civil wars. The results were statistically signi cant for different speci cations and for several robustness checks. The present contribution has succeeded in building a simple theoretical model and performing empirical estimations of the impact of several widelyused welfare state policies on con ict behaviour. However, further research in this area should be encouraged: Generalising the results for a n-player framework would make a lot of sense. Testing the results of the present contribution empirically with micro-level data would also be an interesting research issue. References Agell, Jonas. (996). "Why Sweden s Welfare State Needed Reform", Economic Journal, 06, Alesina, Alberto and Perotti, Roberto. (997). "The Welfare State and Competitiveness", American Economic Review, 87, Alesina, Alberto et al. (2003). "Fractionalization", Journal of Economic Growth, 8, Azam, Jean-Paul. (200). "The Redistributive State and Con icts in Africa", Journal of Peace Research, 38, Baker, Matthew. (2003). "An Equilibrium Con ict Model of Land Tenure in Hunter-Gatherer Societies", Journal of Political Economy,, Brito, Dagobert and Intriligator, Michael. (985). "Con ict, War, and Redistribution", American Political Science Review, 79, Collier, Paul and Hoe er, Anke. (2004). "Greed and grievance in civil war", Oxford Economic Papers, 56, Deininger, Klaus. (2003). "Causes and Consequences of Civil Strife: Micro-level Evidence from Uganda", Oxford Economic Papers, 55, Fearon, James and Laitin, David. (2003). "Ethnicity, Insurgency, and Civil War", American Political Science Review, 97,

20 Freedom House. (2005). Freedom in the World 2004, Washington DC, Freedom House, data le. Gleditsch, Kristian. (2004). "A Revised List of Wars Between and Within Independent States, ", International Interactions, 30: Grossman, Herschel. (994). "Production, Appropriation, and Land Reform", American Economic Review, 84, Grossman, Herschel. (995). "Robin Hood and the Redistribution of Property Income", European Journal of Political Economy,, Grossman, Herschel and Kim, Minseong. (2003). "Educational Policy: Egalitarian or Elitist?", Economics and Politics, 5, Hirshleifer, Jack. (995). "Anarchy and Its Breakdown", Journal of Political Economy, 03, Justino, Patricia. (2005). "Redistribution, Inequality and Political Con ict", Department of Economics, University of Sussex, PRUS working paper 8. King, Gary and Zeng, Langche. (200). "Logistic Regression in Rare Events Data", Political Analysis, 9, Lindbeck, Assar, Nyberg, Sten and Weibull, Jörgen. (999). "Social Norms and Economic Incentives in the Welfare State", Quarterly Journal of Economics, 4, -35. Miguel, Edward, Shanker Satyanath and Ernest Sergenti. (2004). "Economic Shocks and Civil Con ict: An Instrumental Variables Approach", Journal of Political Economy, 2: Muller, Edward and Seligson, Mitchell. (987). "Inequality and Insurgency", American Political Science Review, 8, Noh, Suk Jae. (2002). "Production, Appropriation, and Income Transfers", Economic Inquiry, 40, PRIO (International Peace Research Institute, Oslo). (2006). Data on Armed Con icts, Oslo, PRIO, data le. Rohner, Dominic. (2006). "Beach Holiday in Bali or East-Timor? Why Con ict Can Lead to Under- and Overexploitation of Natural Resources", Economics Letters (forthcoming). Schelling, Thomas. (960). The Strategy of Con ict, Cambridge MA, Harvard University Press. Skaperdas, Stergios. (992). "Cooperation, Con ict, and Power in the Absence of Property Rights", American Economic Review, 82,