How Dictators Forestall Democratization Using International Trade Policy 1

Transcription

1 How Dictators Forestall Democratization Using International Trade Policy 1 Kishore Gawande McCombs School of Business Ben Zissimos 2 University of Exeter Business School February 25th, 2017 Abstract: We examine how a dictatorship uses international trade policy to forestall democratization in response to a world price shock. A country s comparative advantage will determine the trade policy that it uses for this purpose. If the shock is to an export (import) price then the dictatorship must raise (lower) the corresponding export tax (import tariff) to maintain political stability. We find support for these predictions in the data for dictatorships, but not for established democracies whose behavior is predicted by Grossman and Helpman (1994). We find that trade policy setting behavior by illiberal democracies is indistinguishable from that of dictatorships. Keywords. Development, dictatorship, institutions, social conflict, trade policy. JEL Classification Numbers: D30, D74, F11, F13, P16. 1 For useful comments, we are grateful to seminar participants at the University of Oxford. 2 Dept of Economics, University of Exeter Business School, Exeter, EX4 4ST, UK. Tel: b.zissimos@exeter.ac.uk 1

2 1 Introduction There is broad agreement that institutions play a fundamental role in determining economic performance, but if some countries are poor because they have bad economic institutions then why do they not exchange them for better ones? The social conflict view of institutions argues that political institutions play a fundamental role in the determination of economic institutions and efficiency. First developed systematically by North (1981), this view argues that agents who control the state should be modeled as self-interested, choosing institutions that maximize their payoffs. The key political institution, in that it determines which agents control the state, is the form of government: dictatorship or democracy. If the form of government is dictatorship then this will enable an elite, a relatively small set of individuals or social groups, to monopolize political power. Economic development failure is thus characterized as the elite using their political power to structure economic institutions not to support economic development but instead to maximize the rents that they capture (Acemoglu, Johnson and Robinson 2005). The literature on the social conflict view of institutions has developed along two branches. One branch focuses on how the elite choose entry barriers, regulations, and inefficient contracting institutions to protect their economic rents and redistribute resources to themselves (Olson 1982, Krussel and Rios-Rull 1996). Another branch considers the choice of political institution, and specifically why a dictatorship of the elite might allow democratization even though it is not in their interests because it forces them to relinquish control over economic institutions (Acemoglu and Robinson 2000, 2006). An implication of this literature is that, in order to better understand economic development failure, we need to focus not so much on democratization itself but on how a ruling elite forestall democratization. The prior literature on democratization focuses on the role of domestic taxation. The main point made by Acemoglu and Robinson (2000) is that democratization represents a credible commitment by the elite to make transfers to the rest of society on a sufficient scale to defuse a threat of revolution. They argue that through democratization the elite hand over the power to set taxation to the rest of society, and in doing so make a credible commitment to continue payments to the rest even after the threat of revolution has dissipated. Although their paper provides a pathbreaking insight into the purpose of democratization, it does not account for an empirical regularity that has come to light in subsequent research. 1

3 This regularity is that the capacity for domestic progressively redistributive taxation is not installed until after democratization. Aidt and Jensen (2009) show this in the data for a sample of ten Western European countries covering the period Besley and Persson s (2009) framework provides the logic. If the capacity for domestic redistributive taxation provides a relatively efficient mechanism to tax the elite s wealth, then this undermines their incentive to install it: an implication of North s (1981) logic discussed above. Consequently such countries tend to rely instead on relatively inefficient trade taxes for the purposes of redistribution (Besley and Persson 2011). In response to this research, the present paper first develops a theoretical model to consider how dictatorial regimes use trade policy rather than domestic taxation to forestall democratization. Our model builds on Zissimos (2017), which combines Acemoglu and Robinson s (2000) model of democratization with Mayer s (1984) model of international trade policy. A key innovation of our model is that the threat of revolution arises endogenously as a result of a world price shock. Throughout history price shocks, especially food price shocks, have been a key trigger of social unrest, in some cases giving rise to democratization. 3 The present paper first develops a theoretical model to provide a characterization of when a dictator will have to democratize as a way to defuse social unrest, and when they will be able to use trade policy to forestall democratization. In the situation where the elite can forestall democratization, the model delivers concrete predictions as to how trade policy will be used for this purpose. The paper will then test these predictions over how trade policy is operated and find support for them in the data. Our data covers the world food price shocks that have occurred since the 1970s. Thus, although use of international trade policy to forestall democratization has received relatively little attention previously, we show that there appears to be evidence in the data that they have been used for this purpose over the last half century. The structure of the underlying economic model can be understood as follows. Assume a primary-product exporting dictatorship, where production of primary products and manufactures takes place on a competitive basis. The ruling elite own land that is used intensively in the production of primary products. The rest of society rely only on income derived from their labor, which is used intensively in the production of manufactures. According to the Stolper- 3 See Carter, Rausser and Smith (2011) for a review of the literature showing that commodity price volatility can provoke political as well as economic instability. 2

4 Samuelson theorem, elite income from land is maximized when the dictatorship adopts a free trade regime while the rest of society would benefit from the taxation of exports. In our model, over an infinite time horizon, the world relative primary product price (henceforth the world price ) can take one of two levels: low and high. A real income shock comes in the form of a transition to a high world price. If and when that happens, free trade becomes particularly attractive to the ruling elite because returns to land are elevated. But the rest of society will suffer under the high world price, especially if the primary product in question is a food staple that takes up a significant share of their income. If the world price is sufficiently high then at free trade it is worth the rest of society mounting a revolution to overthrow the dictatorial regime so that they can assume power and impose export taxes. We assume that when the rest of society seize power through revolution they democratize. On this basis, while world prices are high, the dictatorial regime have an incentive to put in place export taxes, or increase them if they are already in place, to bring down local prices of the primary product and forestall democratization. This basic logic drives the predictions over the operation of trade policy that we test and find support for in the data. If the country imports the primary product then it is the rest of society that favor a relatively open regime because otherwise the primary product is more expensive at home than on world markets. On the other hand, the elite favor protection of the domestic market for the primary product and, in the process, their land rents. A transition to a high world price creates pressure on the dictatorial regime to reduce domestic protection in order to counter the upward pressure on domestic prices. Surprisingly, an import subsidy on primary products may be required to leave the rest of society indifferent between maintaining the status quo and overthrowing the ruling regime. Again, we take these predictions to the data and find support for them. While the focus of our model is on how dictatorships forestall democratization, the main testable predictions of the model concern trade policy responses to world price shocks. On this basis, it is worth comparing the predictions of our model to those of Grossman and Helpman (1994 henceforth GH), which is the leading model in trade policy formation. GH assume a stable institutional environment where democracy has already been consolidated. In that paper, interest groups are able to lobby the government in order to try to sway policy in the direction they would like to see it go. Because a high world commodity price raises the incomes of commodity producers, a world commodity price shock reduces the incentive at the margin 3

5 to lobby for export subsidies and so they fall. Note that if the export sector is organized then the GH model cannot explain an increase in export taxes because an export lobby would have an incentive to prevent them. On the import-competition side, a world commodity price shock would reduce the incentive at the margin for an import-competing sector to lobby for protection and hence bring about a reduction in import tariffs. In the GH framework, it is not generally worth the import-competing lobby supporting import subsidies because of the distortions entailed. Our empirical implementation covers the period Over that time frame, we associate the world commodity price shocks of and with periods where the world price is high, while in other years it is considered to be at a normal level. The clearest-cut empirical results we obtain are for trade policies on exports so we discuss these first. In line with the predictions of our model, we find that dictatorships respond to the commodity price shocks by raising export taxes. Because the literature on international trade policy has focused on liberal democracies in the past, we use them as a benchmark in our regressions. We find that, in contrast to dictatorships, liberal democracies responded to the world price shock of by reducing export subsidies. The outcome for liberal democracies is exactly what would be predicted by GH. In our dataset, if countries are not classified as dictatorships or liberal democracies then they are classified as middle democracies. Our results show that middle democracies respond to the world commodity price shock of in a way that is indistinguishable from dictatorships: they too increase export taxes. Thus our results show that, from an export-policy standpoint, middle democracies have more in common with dictatorships than they have with liberal democracies. It is worth pausing to consider how we might interpret this result, since one might have expected all democracies to respond in a similar way. In part this outcome might be driven by the fact that the ballot box does not represent an effective means through which power changes hands in some middle democracies, so threats to the ruling regime are addressed in a way that more closely resembles the behavior of dictatorships. Zacharia (1997) discusses the rise of illiberal democracies over recent years. In illiberal democracies, even though power may notionally change hands at the ballot box, to the extent that incumbents are able to curtail such liberties as political representation, it may be possible for them to manipulate policy outcomes in their 4

6 interests in much the same way as under dictatorship. An association of middle democracies in our dataset with Zacharia s illiberal democracies would be consistent with our observation that export policy outcomes in middle democracies are indistinguishable from those in dictatorships. When we turn our attention to import protection, we find that again in line with the predictions of the GH framework the world commodity price spikes did lead to a decrease import tariffs in liberal democracies. But dictatorships, on the contrary, raised import subsidies in primary products especially for staple food crops. While this is at odds with the predictions of GH, it fits naturally with the predictions of our model. This draft is preliminary and incomplete. The next draft will have a full discussion of how our work contributes to the literature. The paper proceeds as follows. Section 2 sets out the underlying economic model, and uses this to motivate the preferred price levels of the elite and the rest of society. Section 3 formalizes the way that a world price shock can endegenously create a threat of revolution, and how the elite will attempt to forestall democratization using international trade policy. It is here that our testable predictions will be developed. Section 4 then introduces the data that will be used to test the predictions, and Section 5 presents the econometric methodology. Section 6 discusses our econometric results. Section 7 concludes. 5

7 2 The Model Consider a single small country that takes world prices as given. The model has an infinite time horizon to capture the commitment problem faced by a ruling elite under the threat of revolution. Production is derived from factors whose endowments are constant. Variation in outcomes over time comes entirely from policy choices, determined by the group that holds power: the ruling elite under dictatorship or the rest of society under democracy. To abstract from issues of domestic taxation, we will assume that there is no domestic fiscal capacity and trade taxes are the only fiscal policy instruments available. 2.1 The Production Structure The production structure is basic Heckscher-Ohlin. There are two homogeneous goods: manufactures and primary products. These are both produced using labor and land. Manufactures are intensive in labor while primary products are intensive in land. Technology exhibits constant returns to scale and decreasing returns to each factor. Goods and factor markets are competitive, and there is free mobility of factors between goods sectors. So in each period t = 0, 1,...,, there is a single price for each factor that reflects its marginal value in either sector: r t for land and w t for labor. There is a continuum of risk neutral agents in the economy, allocated to one of two groups: the ruling elite, ε, and the rest of society, ρ. The population of the elite and the rest of society are normalized to θ and 1 respectively, so the total population is 1+θ. The endowment of labor, 1 + θ, is distributed evenly across the elite and the rest of society. The endowment of land, θ, is distributed evenly but solely among the elite. Therefore, in period t, elite-member factor income is given by yt ε = r t + w t, while rest-of-society-member factor income is given by y ρ t = w t, and aggregate factor income is y t = (1 + θ) w t + θr t. All members of each group are identical to one another, and each group differs from the other only by its factor endowment. The price of good ε relative to good ρ in period t is denoted by p t. Output of good i {ε, ρ} in period t is denoted by x it. Given initial endowments, population shares, production technology, and free entry, outputs and factor prices are determined by p t, so we may write w t = w (p t ), r t = r (p t ), and x it = x i (p t ) in period t. The autarky price of good ε relative to ρ is denoted by p a while the corresponding world relative price is denoted by p w. If the economy 6

8 is open then goods may be traded internationally but factors are not internationally mobile. The key innovation of the present model over Zissimos (2017) is that we will allow for the occurrence of world price shocks over time. While in Zissimos (2017) p w is fixed, in the present paper it fluctuates randomly. With probability 1 κ the world price level is p w L, while with probability κ the world price is p w H, where κ [0, 1] and pw L < pw H. To characterize equilibrium, it will be useful to think of the price level in terms of a state s {L, H}, corresponding to a world price p w {p w L, pw H }. 2.2 Preferences and Policy Instruments Agents j {ε, ρ} are identical in terms of their preferences and discount factor, β (0, 1). Agent j s expected utility at time 0 is: U j 0 = E 0 t=0 ( ) β t u c j εt, cj ρt where E t is the expectations operator conditional on information available at time t. The ( ) instantaneous utility function, u c j εt, cj ρt, is strictly quasi-concave in goods ε and ρ respectively. ( ) Demand functions for agent j are written as c j it = cj i p t, Y j t, where Y j t is total income of agent j and Y t is aggregate income. 4 Aggregate demand for good i, c it, is given by c it = θc ε i (p t, Y ε t ) + c ρ i (p t, Y ρ t ). As mentioned above, we will restrict the set of policy instruments to trade taxes. Since there is no domestic fiscal capacity, trade subsidies are not available. 5 For convenience, and without loss of generality, we will assume that trade policy is applied to good ε. We have not specified the country s comparative advantage so good ε could be either the country s exportable or its import-competing good. For an import, a domestic price above the world price implies an import tariff (and an export subsidy for an export). For an export, a domestic price below the world price implies an export tax (and an import subsidy for an import). Net trade policy revenue for an individual j in period t, is denoted by tr j t. Let m t be the quantity imported of good ε in period t, and m (p t, Y t ) = c ε (p t, Y t ) x ε (p t ) the corresponding import demand function. Then tr t = tr (p t, p w, Y t ) = (p t p w ) m t (p t, Y t ). Trade taxes drive a 4 Formal definitions of Y j t and Y t are provided below. 5 The assumption that domestic fiscal capacity is unavailable focuses attention on trade policy as the sole policy instrument for redistribution. The results do not depend on the fact that trade subsidies are unavailable. 7

9 wedge between the domestic price and the world price, both for consumers and for producers. Note that, in the absence of revenue for trade subsidies, tr t 0: if the country has a comparative advantage in good ρ then m (p t, Y t ) 0 and p t p w ; if it has a comparative advantage in good ε then m (p t, Y t ) 0 and p t p w. 6 Finally, Y t, measured in terms of good ρ, takes the form Y t = y t + tr t. (1) The total income of a member of group j is given by the sum of factor income and revenue collected from trade policy: Y j t distributed between tr ε t and tr ρ t = y j (p t ) + tr j t, j {ε, ρ}. The assumption regarding how tr t is varies across models. For example, typically, when preferences are assumed to be quasi-linear, tariff revenues are redistributed in lump sum across individuals. When preferences are assumed to be homothetic, tariff revenue is distributed across individuals according to their factor income shares, as in Mayer (1984). 2.3 Group Welfare and Preferred Price Level The welfare of a member of group j {ε, ρ} in period t can be measured using the indirect utility function: W j t = W j ( p t, Y j (p t ) ) = W j (p t ), j {ε, ρ}. (2) Assuming that W j (p t ) is strictly concave in p t, each group s preferred trade policy can be determined from (2) and expressed as a unique value of p t. Following this approach, we will say that group j s preferred price level, ˆp j, is the value of p t that maximizes W j t : dw j t /dp t = 0 is the first order condition from which ˆp j is obtained. This solution to ˆp j may imply a trade tax or trade subsidy. If the interior solution implies a trade subsidy then, in the absence of domestic fiscal capacity, the solution will be at a corner: ˆp j = p a or ˆp j = p w. We can see from the above specification that the solution for ˆp j depends on p w more generally. Since ˆp j can be expressed as a function of p w, and since p w {p w L, pw H }, we may write ˆpj s = ˆp j (p w s ), j {ε, ρ}, s {H, L}. We can go one step further and simply write ˆp j s on the understanding that this corresponds to p w s. It will be helpful to impose some further structure on the relationship between world prices and group welfare. We will say that W j (p t ) is Stolper-Samuelson consistent if two properties 6 Exports are denoted by negative imports. 8

10 hold: ˆp ε s > p w s > ˆp ρ s; W ε (ˆp ε H) > W ε (ˆp ε L) while W ρ (ˆp ρ L) > W ρ (ˆp ρ ) H Following Jones (1965), we can express the main implication of the Stolper-Samuelson theorem as r t > p t > 0 > w t, where a superscript- on a variable denotes proportional change, e.g. r t = dr t /r t. Ignoring trade tax revenue and using p w as a benchmark, the Stolper-Samuelson theorem implies that group ε gain from an increase in p t from p w because r increases, while it takes a decrease in p t from p w for group ρ to gain. Therefore, if the Stolper-Samuelson theorem holds, plus W j (p t ) is concaive in p t, all we need for ˆp ε s > p w s > ˆp ρ s is that the share of tariff revenue should be sufficiently small relative to factor income. This establishes the first property directly. In addition, providing the Stolper-Samuelson theorem holds globally, any change in p w will have a monotonic effect on factor incomes. 7 Providing W j (p t ) is montonically increasing in factor incomes, which it will be under standard assumptions, any increase in p w will therefore increase W ε and reduce W ρ, holding ˆp ε and ˆp ρ constant. The effect on W j of any adjustment of ˆp j in response to the shock to p w will be of second order, while the effect of p w on W j will be first order. This establishes the second property. Taking p w s as given, we will now provide a complete characterization of preferred price levels ˆp j s, under a comparative advantage in good ε and ρ respectively. With a comparative advantage in good ε, p w s > p a. In this situation, the elite s welfare is maximized at a higher level of openness than is the rest of society s: with a comparative advantage in good ε, the elite own the factor used intensively producing the good for which the country has a comparative advantage. So the elite s factor income is increasing in openness whereas the rest of society s is decreasing. Note that ˆp ε s > p w s implies an export subsidy for the elite. In the absence of domestic fiscal capacity, the solution to ˆp ε implies a corner solution ˆp ε s = p w s. Turning to ˆp ρ s, if this is at an interior solution then it implies the rest of society would ideally like an export tax on good ε that allows some openness. But, depending on their relative preferences for goods ε and ρ, it could equally be that they would prefer an export subsidy, so in the absence of fiscal capacity the solution to ˆp ρ implies a corner solution at autarky. Therefore, for the rest of society, in the absence of domestic fiscal capacity, p a ˆp ρ s < p w s. 7 The conditions required for this relationship to hold globally are established by Chipman (1969) and the literature to which he refers. These conditions are assumed to hold throughout our analysis. 9

11 With a comparative advantage in good ρ, p a > p w s, in which case the rest of society s welfare is maximized at a higher level of openness than is the elite s. Now the rest of society would ideally like an import subsidy, but in the absence of domestic fiscal capacity must settle for free trade: ˆp ρ s = p w s. The elite may prefer an import tariff that implies some openness. Or they may prefer an import subsidy, in which case in the absence of domestic fiscal capacity they must settle for a corner solution at autarky. Therefore, for the elite, in the absence of domestic fiscal capacity, p w s < ˆp ε s p a. This analysis extends Proposition 1 of Zissimos (2017) to our present more general setting where world price shocks can occur. Proposition 1. In the absence of domestic fiscal capacity, taking p w s as given: (i) with a comparative advantage in good ε, ˆp ε s = p w s while p a ˆp ρ s < p w s ; (ii) with a comparative advantage in good ρ, ˆp ρ s = p w s while p w s < ˆp ε s p a. From this result we can see that factor ownership and comparative advantage co-determine the level of openness preferred by the respective groups. 8 3 World Price Shocks We want to capture the way that the ruling elite set trade policy to forestall democratization, in response to the threat of revolution caused by a world price shock. Initially, (de jure) political power is held by the elite. They exercise their power over policy through a single trade tax: either an export tax or an import tariff, depending on comparative advantage. For parsimony of notation we will say that while the elite hold power they set p t directly, denoting this by p ε t. The rest of society can mount a revolution in any period, and if they do so then it is successful for sure. Revolution is costly, but allows the rest of society to install democracy, through which the median voter determines trade policy, p t. Given θ < 1, the median voter is a member of the rest of society, and so democratization transfers the power to set trade policy from the elite to the rest of society. So democratization implies a transfer of power to set trade policy between the groups. 9 The same outcome of democracy can arise if the elite decide voluntarily 8 The conflict of interest between groups over trade policy is of course more general than the framework we are using here, which is constructred to bring out the revenue considerations of trade policy. 9 Thus, revolution would result only in a transfer of (de jure) political power from the elite to the rest of society. In AR, revolution would also result in a transfer of elite endowments to the rest of society as well. Transferring all elite endowments to the rest of society in a revolution would only affect the range of values of β for which a revolution constraint binds; see the discussion following (5). 10

12 to extend the franchise. Since revolution and franchise extension both lead to democracy, the form of government, is either democracy, D, or elite rule, E. 10 The game is initialized with the assumption that in period 0 there is elite rule. Within a period, t, the sequence of events is as follows. 1. The world price level, p w {p w L, pw H }, is revealed. 2. The elite decide whether or not to extend the franchise. If they do then there is democracy. If they do not, they set trade policy, p t = p ε t. 3. If the elite have not extended the franchise then the rest of society decide whether or not to mount a revolution. If they do so it is successful for sure, leading to democracy. 4. If there is democracy then trade policy p t is set by the median voter (a member of the rest of society). 5. Production takes place, demands are realized, markets clear and consumption takes place. Some additional assumptions are needed to complete the specification of the model. If democracy does not arise in period t, then in period t + 1 the sequence of events starts again at stage 1 and proceeds through all stages. If in period t democracy does arise then in t + 1, p w {p w L, pw H } is determined in stage 1 as before, but stages 2 and 3 are skipped, moving straight to stage 4 where the median voter sets trade policy. 11 The assumption that all members of each of the two respective groups are identical to one another (but obviously differ across groups by their endowments) means that we can model the members of each group as a single player. So the game between the elite and the rest of society can be modeled as a two-player game. From the sequence of events set out above, we can see that the characterization of equilibrium will involve comparing the payoffs of the elite and the rest of society under the status quo, extension of the franchise, and revolution. A challenge with making these comparisons identified by Zissimos (2017) is that ˆp j s may be different under revolution than under the status quo and extension of the franchise. One solution to this issue is to identify features of the model 10 It is straight forward to extend this framework to allow the elite a third option of using repression. Doing so reveals that they repress when it is the cheapest option. 11 The assumption that democracy is an absorbing state enables us to focus the analysis on whether or not it is possible to set trade policy to forestall democratization. Acemoglu and Robinson (2001) present a model where democracy may fail to consolidate, and the present model could straight-forwardly be extended in that direction. 11

13 under which ˆp j s remains constant even in the event of a revolution. For example, Zissimos (2014) shows that when preferences are quasi-linear the choice of ˆp j s is invariant to income and hence the cost of revolution. Alternatively, Proposition 2 of Zissimos (2017) shows that, by characterizing the cost of revolution as a radial contraction of the production possibility frontier, ˆp j s is unaffected by revolution. For notational simplicity, we will assume here that that ˆp j s is unaffected by revolution, as would be the case under the models of of Zissimos (2014) or (2017). This assumption could be relaxed without affecting the analysis, but at the expense of greater notational complexity. We will also assume, as in Zissimos (2014, 2017), that the cost of revolution is borne entirely in the period when the revolution occurs. Denoting a member of group j s income under revolution by Y j R (p t), we will write Y j R (ˆpρ s) = ψy j (ˆp ρ s), where ψ < So the cost of revolution to agent j can be written in terms of income as Y j (ˆp ρ s) Y j R (ˆpρ s) = (1 ψ) Y j (ˆp ρ s). ) Corresponding to this, we will write W ρ R (ˆpρ s) as short-hand for W (ˆp j ρ s, Y j R (ˆpρ s), the expected welfare for a member of group j under revolution. This way of capturing the payoff to revolution will be useful in the characterization of equilibrium. 3.1 Definition of Equilibrium, Payoffs, and the Commitment Problem The approach to characterization of equilibrium follows Zissimos (2017). The concept of equilibrium is Markov Perfection, wherein each player s strategy depends only on the state (F ; s) in a given period. The strategies played by the respective groups are as follows. The strategy σ ε (F ; s) played by the elite consists of the choice over whether or not to extend the franchise, and how to set trade policy. Let f = 0 if the elite do not extend the franchise (in which case p ε t is their choice of trade policy) and f = 1 if they do. Let σ ρ (F (f, p ε t) ; s) be the strategy played by the rest of society in response to the choices f and p ε t by the elite. This consists of the rest of society s decision as to whether or not to mount a revolution: a = 1 if they do (where a is a mnemonic for agitate ) and 0 otherwise. Since, by the timing of events determined above, the elite move before the rest of society, the strategy of the rest of society in a given period is conditioned on that of the elite. Let σ ε (F ; s) be a best response to σ ρ (F (f, p ε t) ; s) for all F, s, and let σ ρ (F (f, p ε t) ; s) be a best response to σ ε (F ; s) for all F, s. Then a pure strategy Markov Perfect equilibrium is a set of mutual best responses { σ ε (F ; s), σ ρ (F (f, p ε t) ; s)}. 12 Using the notation we developed earlier, when p t = ˆp ρ s, Y j (ˆp ρ s) denotes agent j s income when there is no revolution. 12

14 To examine which outcome will arise in equilibrium, we will now formalize the payoffs to the respective groups under the various possible outcomes. Let V j (D, ˆp ρ s) represent the present discounted value under democracy for j {ε, ρ}. democracy via an extension of the franchise takes the form: V j (D, ˆp ρ s) W j (ˆp ρ s) + For a member of group j, the payoff to βκ 1 β W j (ˆp ρ ) β (1 κ) H + 1 β W j (ˆp ρ ) L. (3) The fact that the first term depends on ˆp ρ s allows for the possibility that in the current period s is either H or L. The weight on the second term, βκ/ (1 β), provides the net present value of W j (ˆp ρ H), weighted by the fact that the probability that this arises over time is given by κ. The weight on the third term, β (1 κ) / (1 β) provides the same thing for W j (ˆp ρ L), given a corresponding probability of 1 κ. Denoting the occurrence of revolution by R, the payoff to revolution is given by V j (R, ˆp ρ s) W j R (ˆpρ s) + βκ 1 β W j (ˆp ρ ) β (1 κ) H + 1 β W j (ˆp ρ ) L. (4) where the first term captures agent j s payoff in the period of revolution, while the second and third terms capture the discounted payoff in subsequent periods. Clearly, given that the continuation payoffs are the same under extension of the franchise and revolution, both groups would prefer an extension of the franchise because this avoids the cost of revolution (1 ψ) Y j (ˆp ρ s). Therefore, extension of the franchise always has the potential to defuse revolution. But the elite can use trade policy to defuse the threat of revolution if the rest of society s expected payoff under trade policy set by the elite is at least as high as under revolution. We are now ready to formalize the commitment problem. The following revolution constraint provides a necessary (but not sufficient) condition for the elite to face a commitment problem. In state s {H, L}, the revolution constraint is binding if: W ρ R (ˆpρ s) + βκ 1 β W ρ (ˆp ρ ) β (1 κ) H + 1 β W ρ (ˆp ρ ) L > W ρ (ˆp ρ s) + βκ 1 β W ρ (ˆp ε H) + β (1 κ) 1 β W ρ (ˆp ε L). (5) The elite will never face a commitment problem if they can make sufficient transfers within a single period to more than compensate the rest of society for what they would obtain from revolution. By (4), the first line of (5) gives the payoff to revolution for the rest of society: V ρ (R, ˆp ρ s). The second line gives what we call the best today, worst future for the rest of 13

15 society. This gives the payoff to their preferred policy today given the state s {H, L}, followed by the elite s preferred policy in every period in the future, weighted by the expected occurrence of s = H and s = L respectively. Assume a given state in the current period: either s = H or s = L. And assume as abovet that in future periods s = H with probability κ while s = L with probability 1 κ. By construction, W ρ R (ˆpρ s) < W ρ (ˆp ρ s) while in future, whether s = H or s = L, W ρ (ˆp ρ s) > W ρ (ˆp ε s). Therefore, for given cost (1 ψ) Y j (ˆp ρ s), the revolution constraint fails for β sufficiently small and binds for β sufficiently large. That is, a sufficiently large value of β puts a sufficiently large value on future occurrences of W ρ (ˆp ρ s) > W ρ (ˆp ε s). Intuitively, for the revolution constraint to bind, the rest of society must care enough about the future that they obtain a higher payoff from revolution despite the cost. This would happen because they are able to set more favorable trade policy ˆp ρ s over time, than they would from a single period of their best trade policy today, ˆp ρ s, followed by their worst future ˆp ε s, s {H, L}. If the revolution constraint fails, then the elite can induce a higher level of welfare for the rest of society by setting ˆp ρ s for a single period than they could obtain through revolution. Therefore, the elite could in fact use trade policy to set a price p ε t > ˆp ρ s that equates the two sides of (5) for a single period to completely defuse the threat of revolution. Then the commitment problem does not arise because the elite can completely defuse the threat of revolution within a single period. The above characterization of the revolution constraint and commitment problem applies both to s = H and to s = L. Although it is possible that the elite always faces a commitment problem, a potentially more interesting and realistic scenario is one where the elite only faces a commitment problem in periods where s = H. So it is the occurrence of high world prices that puts the elite in a position where they face the threat of revolution and must respond to it using trade policy. Rewriting (5), and assuming for illustration that κ = 0, the condition for the revolution constraint to bind only when s = H can be expressed as W ρ (ˆp ρ ) H W ρ ρ ) β ( R (ˆp H < W ρ (ˆp ρ ) L W ρ (ˆp ε 1 β L) ) < W ρ (ˆp ρ L) W ρ R ρ ) (ˆp L The first term gives the instantaneous gain from maintaining the status quo when s = H, while the third term gives the same thing for s = L. The second term gives the present discounted value of revolution from next period onwards. By our formalization of the cost of revolution, we can write the relationship between the first and third terms as (1 ψ) W ρ (ˆp ρ H) < (1 ψ) W ρ (ˆp ρ L), which always holds under our assumption that W ρ (ˆp ρ H) < W ρ (ˆp ρ L). It then follows immediately 14 (6)

16 that we can set β sufficiently large that the second term is larger than the first, but not so large that it is larger than the third. What about values of κ (0, 1]? Our approach will be to show that the same basic approach as for κ = 0 can be applied to κ = 1, and hence applies for all κ (0, 1) as well. For κ = 1, (6) is exactly the same except the part of the second term in parentheses takes the form ( W ρ (ˆp ρ ) H W ρ (ˆp ε H )). So once again, there exists an intermediate range of β for which this is larger than the first term but smaller than the third. For κ (0, 1), the second term is an average of ( W ρ (ˆp ρ L) W ρ (ˆp ε L )) and ( W ρ (ˆp ρ H) W ρ (ˆp ε H )) weighted by κ. Note that the first and third terms do not vary with κ. It follows that for any given value of κ [0, 1], there exists an intermediate range of β for which (6) holds. Note that the smaller is ψ, the larger are the first and third terms, and so the larger the range of β for which (6) holds. We have now proved a result that establishes the conditions under which the revolution constraint binds only when s = H. Proposition 2. Fix values of p w L, pw H, and ψ. Then there exists a range of values of β sufficiently large that the revolution constriant binds for p w = p w H, but sufficiently small that the revolution constraint does not bind for p w = p w L. Assume a value of β in the range where the revolution constraint binds in periods when when p w = p w H, but not when pw = p w L. It is when the elite face a revolution constraint that they cannot make sufficient transfers within a period to compensate the rest of society for their gains from revolution; they must set trade policy favorable to the rest of society for multiple periods in order to do so. The question is whether they face a commitment problem, and hence must extend the franchise in order to make a credible commitment to transfers sufficient to compensate the rest of society for what they could gain under revolution, or whether they can make sufficient transfers using trade policy over multiple periods and hence use trade policy to forestall democratization. Before we can characterize the equilibrium outcome when p w = p w H, we need to be clear about the outcome when p w = p w L. To simplify the situation in an intuitively appealing way, we will restrict the parameter space to a situation where, under p w = p w L, the elite simply set their preferred policy, ˆp ε L. This requires an additional restriction on the relationship between pw L and p w H, and also a restriction on ψ. Instead of (6), we require 15

17 W ρ (ˆp ρ ) H W ρ ρ ) β ( R (ˆp H < W ρ (ˆp ρ ) L W ρ (ˆp ε 1 β L) ) < W ρ (ˆp ε L) W ρ ρ ) R (ˆp L To obtain this, we have taken (6) and replaced W ρ (ˆp ρ L) with W ρ (ˆp ε L ). Since W ρ (ˆp ε L ) < W ρ (ˆp ρ L), we require ψ to be sufficiently small as to make the third term positive. Then, since W ρ (ˆp ε L ) is decreasing in the world price, the third term will be greater than the first term given a sufficiently large difference between p w L and pw H. The specific conditions are given in the proof of Proposition 2. With the third term larger than the first, as for (6), there exists a range of β for which the second term lies between the two. In that case, when p w = p w L, the instantaneous gains for the rest of society from maintaining the status quo are greater than the continuation gains from revolution, even when the elite set their preferred policy, ˆp ε L, while under pw = p w H the revolution constraint binds. The analysis is summarized as follows. (7) Proposition 3. Fix p w H sufficiently large relative to pw L as to make W ρ (ˆp ε L ) > W ρ (ˆp ρ H). Then there exists a value of ψ sufficiently small that (7) holds. Also, there exists a range of values of β sufficiently large that whenever p w = p w H, the revolution constriant binds, but sufficiently small that whenever p w = p w L, not only does the revolution constraint fail to bind, but it is a dominant strategy for the elite to set ˆp ε L. Proposition 3 is like Proposition 2 except that the benefit to the rest of society from maintaining the status quo is smaller because W ρ (ˆp ε L ) < W ρ (ˆp ρ L). The upper bound on ψ basically ensures that revolution is sufficiently costly in terms of lost output. This ensures that the instantaneous gain from maintaining the status quo is large enough to overcome the continuation gains from revolution even when the elite set ˆp ε L, given a sufficiently large world price shock. This set-up implies that the elite only need to worry about setting policy differently from their preferred level, ˆp ε s, when s = H. Assume from now on that we choose a set of parameter values consistent with Proposition 3, so that the revolution constraint binds when s = H but not when s = L. The question of whether or not the elite face a commitment problem when s = H comes down to whether the elite can set a status quo price p sq [ˆp ρ H, ˆpε H ) in every occurrence of H to yield exactly the same payoff for the rest of society as under revolution. If they cannot then they face a commitment problem. To evaluate this, we need to be able to calculate agent j s payoff under p sq : V j (E, p sq ; H) W j (p sq ) + β ( κv j (E, p sq ; H) + (1 κ) V j (E, ˆp ε L; L) ). (8) 16

18 The first term on the right hand side of (8) shows the payoff to agent j in the current period under p sq. The second term shows the expected payoff in the following period, which is discounted by β. If (with probability κ) in the following period the state of H is maintained then the elite will continue to set p sq and agent j s utility will be maintained at the same level. But if (with probability 1 κ) the state switches to L then the elite will (renege on any promise to maintain redistribution with p sq and) instead implement their preferred price level ˆp ε L. Solving recursively, V j (E, p sq ; H) is then given by V j (E, p sq ; H) = 1 β (1 κ) W j (p sq β (1 κ) ) + 1 β 1 β W j (ˆp ε L) (9) The first term on the right hand side of this recursive solution weights W j (p sq ) by the expected frequency of state H, and the second term similarly weights W j (ˆp ε L ) by the expected frequency of state L. So the sum of these terms gives the expected payoff to agent j from maintaining the status quo. To find p sq, first define the function G (p ε H) V ρ (E, p ε H; H) V ρ ( R, ˆp ρ ) H (10) where p ε H replaces psq in (9). Then, by definition, p sq solves G (p sq ) = 0. This specification formalizes the idea that to maintain the status quo the elite muar set p sq so that the rest of society are just indifferent between mounting a revolution and not doing so. Now that we have specified the commitment problem and payoff functions, we can complete the formalization of the game. According to the above definition of equilibrium, given H, the elite first decide whether to extend the franchise, f = 1, or whether to use trade policy to maintain the status quo, f = 0 and p ε t = p sq. Given either of these two actions by the elite, a member of the rest of society s strategy, a = 0 or a = 1, solves the problem max { fv ρ ( D, ˆp ρ H) + (1 f) V ρ (E, p sq ; H), V ρ ( R, ˆp ρ H)}. (11) The rest of society obtain the first term in brackets if they choose a = 0 and the second term if they choose a = 1. With the game formalized in this way, the characterization of equilibrium is straightforward. This involves evaluating whether or not the elite face a commitment problem. If they do not, they can use trade policy to avoid a revolution (i.e. set f = 0) and if they do then they must extend the franchise in order to do so (i.e. set f = 1). 17

19 3.2 Characterization of Equilibrium To characterize equilibrium, we will now take the first step of examining when the elite face a commitment problem and hence must extend the franchise. They face a commitment problem if, in state H, it is not feasible for them to use trade policy to maintain the status quo. To examine feasibility, let Ṽ ρ (E κ; H) be the maximum utility that the elite can induce for the rest of society using trade policy (as an alternative to extending the franchise). This is induced by setting ˆp ρ H in every period where H arises, and setting ˆpε L in every period where L arises: formally, by setting p sq = ˆp ρ H in (9), Ṽ ρ (E κ; H) V ρ ( E, ˆp ρ H ; H). Then the condition for the elite to face a commitment problem is Ṽ ρ (E κ; H) < V ρ ( R, ˆp ρ H). Building on the approach developed by Zissimos (2017), the next result establishes that there exists a critical level of κ, denoted κ, at which the elite are just able to prevent a revolution using trade policy; for κ < κ the elite face a commitment problem and for κ > κ they do not. Lemma 1. Assume values of β, p w H, pw L and ψ for which the revolution constraint binds for s = H but not for s = L. Then in state s = H there exists a (unique) κ (0, 1) at which Ṽ ρ (E κ; H) = V ρ ( R, ˆp ρ H). 1. For all κ < κ, Ṽ ρ (E κ; H) < V ρ ( R, ˆp ρ H) : over this range of κ it is not feasible to use trade policy to prevent a revolution so the elite face a commitment problem. 2. For all κ > κ, Ṽ ρ (E κ; H) > V ρ ( R, ˆp ρ H) : over this range of κ it is feasible to use trade policy to prevent a revolution so the elite do not face a commitment problem. Figure 1 illustrates this result. The horizontal axis shows the value of κ. For s = H and any value of κ [0, 1], the vertical axis shows the present discounted value to the rest of society from an extension of the franchise, revolution, and the maximum level of expected welfare that the elite can induce for the rest of society using trade policy. The payoff to democracy is calculated by (3) and illustrated by the downward-sloping dashed line. The downward-sloping solid line whose slope is the same as that for the payoff to democracy, shows the payoff to revolution, V ρ ( R, ˆp ρ H), given by (4). The (constant) vertical difference between this and the payoff to democracy is proportional to the cost of revolution, (1 ψ) Y ρ (ˆp ρ H). The reason that the lines are downward sloping is because a higher value of κ increases the likelihood of a high world price, which implies lower expected welfare for the rest of society. 18

20 The solid line whose negative gradient is lower than the other two lines shows Ṽ ρ (E κ; H). The intercept of Ṽ ρ (E κ; H) with the vertical axis, where κ = 0, is Ṽ ρ (E 0; H) = W ρ (ˆp ρ ) β H + 1 β W ρ (ˆp ε L). This is the payoff given by the second line of the revolution constraint, (5), for κ = 0. Recall that we have assumed a value of β at which the revolution constraint binds for s = H. This implies that the intercept of Ṽ ρ (E κ; H) is lower than V ρ ( R, ˆp ρ H), and thus shows graphically that the revolution constraint is a necessary condition for the commitment problem. The endpoint of Ṽ ρ (E κ; H), where κ = 1, is Ṽ ρ (E 1; H) = 1 1 β W ρ (ˆp ρ ) H, which corresponds to the payoff to democracy when κ = 1. If κ = 1 then it is feasible for the elite to set p ε t = ˆp ρ H in every period in the future, inducing the same level of welfare as democracy via an extension of the franchise. By inspection, Ṽ ρ (E 0; H) > Ṽ ρ (E 1; H) and so Ṽ ρ (E κ; H) must be downward sloping. The fact that the gradient of Ṽ ρ (E κ; H) is smaller than for V ρ ( D, ˆp ρ H) arises from the fact that an increase in κ reduces the expected occurrence of W ρ (ˆp ρ ) L in the case of democracy, but reduces the expected occurrence of W ρ (ˆp ε L ) in the case of the status quo; all other terms in the gradients are the same. Since W ρ (ˆp ρ L) > W ρ (ˆp ε L ), expected welfare must decline more quickly with an increase in κ under V ρ ( D, ˆp ρ H) than under Ṽ ρ (E κ; H). Given this structure, there must exist a value, κ, at which Ṽ ρ (E κ; H) = V ρ ( R, ˆp ρ H). The fact that the payoffs to extension of the franchise, revolution, and the maximum feasible welfare that the elite can induce for the rest of society are all decreasing in κ represents a striking difference in the present framework to that of Zissimos (2017). There, the payoffs where constant in κ for extension of the franchise and revolution, and increasing in κ for the rest of society s maximum feasible welfare under the status quo. The key difference is that here the low threat state is associated with a low world price, and hence welfare tends to be higher for the rest of society. Of course, the trade policy set by whichever group is in power may exacerbate or mitigate this underlying tendency. This feature was missing from Zissimos (2017), with policy being the only cause of variation in welfare. Now consider the elite s options in state H. For κ > κ, it is feasible for the elite to use trade policy to maintain the status quo because they can feasibly induce a level of welfare in the rest 19

21 of society that is at least as great as from revolution: Ṽ ρ (E κ; H) > V ρ ( R, ˆp ρ H) over this range. This is based on the fact that the expected number of periods in the future for which the elite can set ˆp ρ H as opposed to ˆpε H is sufficiently large. For κ < κ, Ṽ ρ (E κ; H) < V ρ ( R, ˆp ρ H) because the expected number of periods in the future for which the elite can set ˆp ρ H is not sufficiently large. So in this case the elite do face a commitment problem, and must extend the franchise in order to defuse the threat of revolution. Having established the range of κ for which it is feasible for the elite to use trade policy to maintain the status quo, we will now examine the trade policy that the elite actually set in equilibrium, by providing a characterization of p sq. Letting p ε H = ˆpε H in (10), G (ˆpε H ) < 0 by the revolution constraint. Therefore, if a value p sq exists such that G (p sq ) = 0, it must be because the payoff in the current period under p sq is W ρ (p sq ) > W ρ (ˆp ε H ). And by the optimality of the choice of ˆp ρ H, it must be the case that W ρ (ˆp ρ H) > W ρ (p sq ). Putting these observations together, W ρ (ˆp ρ H) > W ρ (p sq ) > W ρ (ˆp ε H ). It also follows, by the fact that ˆpε H > ˆpρ H and the concavity of W j (p t ), that ˆp ε H > psq > ˆp ρ H. Note, again by the concavity of W j (p t ), that there is another value of p t that would yield the same level of welfare for ρ as p sq, but the elite would never choose this because it is even further away from ˆp ε H than ˆpρ H. We can therefore also say that W ε (ˆp ε H ) > W ε (p sq ) > W ε (ˆp ρ H). The final step is to prove that a value of p sq exists such that G (p sq ) = 0. Recall that, by definition, G (ˆp ρ H) = 0 at κ. And by inspection of (9), V ρ (E, p sq ; H) is increasing in κ for given p sq. So G (ˆp ρ H) > 0 for κ > κ and, by concavity of W ρ (p t ) in p t, there must exist a unique value of p sq > ˆp ρ H at which G (psq ) = 0. Moreover, p sq increases continuously as κ is increased. But it cannot reach ˆp ε H by the fact that the revolution constraint is binding. So there must exist a value of p s for which ˆp ε H > psq > ˆp ρ H. We have now proved the following result: Proposition 4. Assume values of β, p w H, pw L and ψ for which the revolution constraint binds for s = H but not for s = L. In state s = H, for κ > κ, there exists a unique status quo price, p sq, that entails a compromise between the two groups in the sense that: (i) ˆp ε H > psq > ˆp ρ H ; (ii) W ρ (ˆp ρ H) > W ρ (p sq ) > W ρ (ˆp ε H ) ; and (iii) W ε (ˆp ε H ) > W ε (p sq ) > W ε (ˆp ρ H). This result extends Proposition 3 of Zissimos (2017) to the present setting. It shows that we can characterize the status quo price strictly in terms of the world price and corresponding preferred price levels, p sq, ˆp ε H and pρ H, strictly in terms of outcomes in state s = H. Having 20

22 pinned down the policy outcome in s = L to one where the elite plays a dominant strategy of maintaining power and setting p ε t = ˆp ε L, the characterization of equilibrium is reasonably straight forward, as established in the next result. Proposition 5. Assume parameter values as in Proposition 4. For κ κ there exists a unique pure strategy Markov Perfect Equilibrium with the following characteristics. For any value of κ κ, if p w = p w L, the elite face no threat of revolution and adopt their preferred price level, ˆp ε L = pw L. (i) If κ < κ and if p w = p w H extending the franchise. (ii) If κ > κ and if p w = p w L then the elite do face a threat of revolution and will respond by then the elite do face a threat of revolution and respond to the threat by temporarily raising the rest of society s welfare by setting the status quo price, p sq, using trade policy. This results extends Proposition 4 of Zissimos (2017) to the present setting. It shares with the earlier result the feature that a high threat state requires the elite to respond in order to avoid a revolution. In state s = H, the elite can credibly set p sq instead of ˆp ε H. Since W ρ (p sq ) > W ρ (ˆp ε H ), setting the status quo price has the potential to defuse the threat of revolution. In case (i), κ < κ the high threat state simply is not expected to arise sufficiently frequently for the elite to use trade policy to make transfers to the rest of society, and the elite do face a commitment problem. Hence they must extend the franchise. In case (ii), κ > κ and so the elite have more opportunities to raise the rest of society s welfare by setting p sq. As a result, the elite do not face a commitment problem, and so they can use trade policy to maintain the status quo. There is, however, a key difference between the present result and Proposition 4 of Zissimos (2017). As mentioned above, the difference between the present setting and Zissimos (2017) is that here the low threat state is associated with a lower world price and hence higher welfare for the rest of society. In Zissimos (2017), policy is the only source of variation in welfare, and the low threat state is associated with a low level of welfare because in that state the elite can set with impunity a policy that is bad for the rest of society. In the present setting, differently from Zissimos (2017), there are two sources of variation: the world price level itself, and the policy response to the world price level. Here, differently from Zissimos (2017), the rest of society s 21

23 welfare may be higher in s = L than in s = H because the world price level is more favorable to them. The fact that the rest of society s welfare may be lower in s = H than s = L turns out not to change the basic logic of the argument. The logic of the argument rests, instead, on the elite being able to induce an increase in welfare, W ρ (p sq ) > W ρ (ˆp ε H ), by setting psq instead of ˆp ε H with sufficient frequency, i.e. κ > κ. Observe that the above characterization of equilibrium says nothing about the good for which the country has a comparative advantage. The next subsection treats the case of a comparative advantage in good ε, and the one after that considers a comparative advantage in good ρ. 3.3 Comparative Advantage in Good ε Recall that, without loss of generality, trade policy is applied to good ε. So with a comparative advantage in good ε, the trade policy is either an export tax or an export subsidy. We will obtain clear indicative predictions by restricting attention to a situation where there is no domestic fiscal capacity. Recall from Proposition 1(i) that, in the absence of fiscal capacity, with a comparative advantage in good ε, ˆp ε = p w while p a ˆp ρ < p w. In other words, the elite s preferred price level implies free trade while that of the rest of society implies an export tax. We will use this result in conjunction with the characterization of equilibrium that we obtained in Propositions 4 and 5. Also, the discussion will focus on the situation where the elite do not face a commitment problem and so can use trade policy to maintain the status quo. Accordingly, we restrict the parameter space to the range where κ > κ. Now let us consider the policy outcomes under p w L and pw H respectively. We know from the characterization of equilibrium in Proposition 5 that: if p w = p w L, the elite set ˆpε L = pw L. In other words, they adopt free trade; if on the other hand p w = p w H then while the elite s preferred price level is ˆp ε H = pw H, they must set the status quo price, psq, in order to defuse the threat of revolution, and we know from Proposition 4 that ˆp ε H > psq > ˆp ρ H. With group j {ε, ρ} in power, the formula for an ad valorem export tax set in state s takes the form τ EX s = ( ) ˆp j s p w s /p w. Using the equilibrium values determined above, we therefore have τ EX L = 0 and τ EX H = (psq ˆp ε H ) /ˆpε H < 0, which implies an export tax. Proposition 6. Assume parameter values as in Proposition 4, κ > κ, and a comparative 22

24 advantage in good ε. There exists a unique pure strategy Markov Perfect Equilibrium with the following characteristics: if p w = p w L, then the elite face no threat of revolution and so adopt their preferred trade policy of free trade, τ EX L = 0; if p w = p w H then the elite do face the threat of revolution and defuse this by setting p sq using an export tax, τ EX H = (psq ˆp ε H ) /ˆpε H < 0. So we see from this that, with a comparative advantage in good ε, trade policy will switch between free trade and an export tax, according to whether p w = p w L or pw H. This is the prediction that we test for in the data. 3.4 Comparative advantage in good ρ To be added 23

25 4 Data Dependent Variable The dependent variable in our analysis is the nominal rate of assistance, or NRA, afforded by governments to exported agricultural products. Anderson and Valenzuela s (2008) ambitious measurement of NRAs spans several countries over the six decades The nominal rate of assistance for a farm product is the percent increase, due to government policies, in gross returns to farmers above what they would have been without the government s intervention. Taking the world price as a reference, a product s NRA is: positive when government raises the product s price above the world price; negative when government lowers the product s price below the world price. A positive NRA on export products amounts to subsidizing the product, while a negative NRA amounts to taxing it. The NRA price distortions measures are theoretically encompassing, and include a range of instruments: domestic and trade policy instruments, including border price supports, exchange rate distortions, production subsidies and taxes, and input price distortions (Anderson et al. 2008). Hovewever, the predominant distorting influences contained in the NRAs are border distortions distortions intended to distance domestic price from world price not domestic distortions (Anderson et al. xx). This is especially true in developing countries. Suppose India subsidized its exports of rice with an ad valorem subsidy s x. It would be precisely measured by the NRA, operationally defined as NRA = E P (1 + s x) E P E P, (12) where E is the Indian rupee (INR) per U.S. dollar ($) rate and P is the INR price of rice on the international market. Taking this formula to the field to compute this NRA requires information about the domestic INR price received by the Indian rice producer at the farmgate, P (1 + s x ), as provided by Anderson et al. (2008). If the subsidy is the sole distortion, the NRA on rice exports computes to s x > If rice exports are taxed overall then s x < 0. A subsidy (positive NRA) assists rice farmers, while an export tax (negative NRA), hurts rice farmers by redistributing income away from rice farmers. The classic example is a government-imposed ban on exports of a commodity, whose ad valorem equivalent may be computed using (xx) once 13 The formula flexibly allows India to be a small rice exporter (its actions do not affect world rice price, or E P = P ) or a large rice exporter (its rice subsidy pushes down world prices). 24

26 field work determines the farmgate price P (1 + s x ). Anderson et al. properly account for exchange rate distortions, often used by developing countries well into the mid-1980s, partly as a redistributive policy instrument. Domestic prices in the study are converted to U.S. dollars using market foreign exchange rates, or multi-tiered exchange rates, or shadow exchange rates estimated in other studies, to take into account the distortions to the foreign exchange market. 14 Input, and other supply chain, distortions are also accounted in their computations. Details of the methodology may be found in Anderson (2009, Appendix A). Our primary dependent variables are overall NRAs for a country s exports, computed by Anderson and Valenzuela as the trade-weighted averages of that country s NRAs on exported products. Regressors Since the models include country fixed-effects, we use two time-varying regressors as control variables: log per capita GDP and log nominal exchange rate. A key variable in the analysis is land inequality measured by the Gini coefficient of landholdings. We defer its detailed description to the section below. 5 Econometric Models We begin with models that test the simpler hypotheses, not specific to non-democracies, that price spikes will be met with export taxes or a reduction in export subsidies. From models with country-fixed effects (FE model) and important control variables (described below), we estimate the difference between pre-shock and post-shock NRAs for the sample of non-democracies. We then do the same for the full sample, where we differentiate between non-democracies, middle democracies and liberal democracies. The second model we use is first-differenced (FD model), 14 Where a country has distortions in its domestic market for foreign currency, the exchange rate relevant for calculating the NRA for a particular tradable product depends, in the case of a dual exchange rate system, on whether the product is an importable or an exportable, while in the case of multiple exchange rates it depends on the specific rate applying to that product each year (Anderson et al. 2008, p. 684). 25

27 which eliminates fixed effects. The FE model for export NRAs is NRA i,t = αnra i,t 1 + βi HIGH t + X i,t B + γ i + e i,t, (13) where NRA i,t is country i s overall export NRA in year t, γ i are the fixed effects, and panel dynamics are accounted by the lagged dependent variable NRA i,t 1. The vector X i,t includes three control variables: (log) per capita GDP, (log) exchange rates and a linear trend. Conditional on the regressors, we presume the error term e i,t to be iid normal. We conservatively estimate the parameter estimates with robust standard errors adjusted for clustering by country to account for within-country correlation in NRAs over time. The issue of interest is the coefficient β on the indicator variable It HIGH, which equals one for the years during which agricultural price spiked. We analyze the effect of spikes in two very different eras: the spike and the spike. The spike was the first sharp increase in food prices in the post-war era. The spike coincided with a period of macroeconomic shocks, first with the collapse of the Bretton Woods system of fixed exchange rates, followed by oil price shocks initiated by the OPEC cartel. The sample period for this analysis comprises the period to include years prior to and following the spike. Non-democracies in this sample are quite different from non-democracies in the second sample period of (when the availability of NRA data end) which span years around the spike. Through the 1980s and 1990s, the price of agricultural products entered into a long decline, despite the spike in The spike is significant because it brought prices back, in real terms, to those that prevailed before the 1970s spikes (Sumner 2009, Fig. 3). Our hypothesis, which follows from the Proposition 6 is that β < 0. While not part of the theory, it makes intuitive sense that regardless of whether the political leadership were elected or selected into power, a form of the revolution constraint in (xx) would apply: In democracies, the fear of losing power at the booth would incentivize lower food prices in the face of high price shocks. We therefore present estimates of β for non-democracies, middle democracies and liberal democracies from the model, NRA i,t = φnra i,t d=1 α d (DEMd i I HIGH t ) + X i,t B + γ i + e i,t, (14) where DEM1 i = 1 if country i is a non-democracy, DEM2 i = 1 if i is a middle democracy, and DEM3 i = 1 if i is a liberal democracy. These are defined for each country in the sample 26

28 on the basis of their Polity score (Marshall et al. 2013). Polity scores measuring the quality of democracies and non-democracies, were introduced by Marshall et al. (1992), and remain widely used. Our three democracy categories are measured for each country in the sample on the basis of its ten-year average Polity score preceding the start of the sample period. For the price shock, we use their for Polity scores vary from -10 to +10 (more on this below xxx footnote it here, and why Freedom House is not used). A country is assigned to type DEM1 (non-democracy) if its average score is between -10 and 0, to type DEM2 (middle-democracy) if its average is between 1 and 8, and to type DEM3 (liberal democracy) if the average is 9 or a perfect 10. We posit that regardless of the type of regime, during a high price shock, trade policy will attempt to ease domestic prices, that is, α d < 0, d = 1, 2, 3. While these models can provide evidence about whether trade policies are altered in the face of food price shocks, they are less informative about the hypothesis that is core to our theory: that the elite concede export taxes on products that they gain from freely exporting or subsidizing (if they could). The proper test of this core hypothesis requires a specification recognizing the elite as land holders. In countries with the greatest land inequality, either the elite in political power are also largeholder, or powerful largeholders are connected with the elite because they protect the interests of these largeholders (Binswanger et al in Asia and Africa; Goldstein and Udry 2008 in Ghana; de Ferranti et al (Ch. 6) in Latin America). Our main results are therefore from the model: NRA i,t = φnra i,t d=1 α d (DEMd i I HIGH t ) + 3 d=1 β d ( DEMd i [ LGINI i It HIGH ]) + X i,t B + γ i + e i,t, (15) where LGINI i is land inequality in country i. Countries that confer large-holdings to the elite and small-holdings to the rural masses are have high land Gini measures. Deininger and Squire (1998) seminally constructed their measure of inequality of holdings among landholders for countries. Vollrath (2007) adds more countries to that database. We complete the land Gini data with entries from the Food and Agricultural Organization s compilation based on their landholdings surveys of 2007 (FAO 2007) as well as individual sources documented in the Data Appendix. From these sources LGINI is measured for the cross-section of countries around Where Deininger and Squire do not provide a measure, we use individual sources as 27

29 described in the data appendix. 15 These authors compute land Gini coefficients using data on the size distribution of land holdings obtained from the decennial censuses of the FAO. Data on the total number of holdings and total area of holdings, by size of holdings, is then used to estimate a Lorenz curve from which the land Gini is calculated. The Ginis are comparable across countries. We note that they measure the distribution of operational landholdings within a country (Vollrath, 2007). The hypotheses we test are: (i) β 1 < 0, and (ii) β 2 = β 3 = 0. Not only does our theory predict that the elite, who have much to gain from free trade in products that use land that the elite own abundantly intensively, will impose taxes and prevent trade in those goods during periods experiencing high price spikes, but that the land owning elite in middle and liberal democracies will not behave in the same way. Elected governments in democracies will have the same incentives to reduce prices by imposing taxes (or reducing subsidies) on food exports, but there is no compelling reason why it will be related to the elites landholdings. We also report estimates of the same models using first-differences (FD) rather than fixed effects to remove the unobserved effects. The FD counterpart to the model (xx) is: NRA i,t = φnra 0 + α I HIGH t + X i,t B + e i,t, (16) where z z t z t 1. The pre-shock NRA, NRA 0, captures whether countries are even able change NRAs. If NRA 0 > 0 (subsidized) or NRA 0 is negative but small (low taxes), there is greater latitude to impose further taxes than if the pre-shock export tax were already high or prohibitive. Therefore, we expect α < 0. We test the main prediction of the theory from the FD model, as well, with land Gini interactions: NRA i,t = φnra d=1 α d (DEMd i I HIGH t ) + 3 d=1 β d ( DEMd i [ LGINI i It HIGH ]) + X i,t B + e i,t. (17) The FE and FD models with the land Gini interactions are somewhat distinct in the following sense. While the FE model uses within variation (using all available sample data points within 15 For example, China s rural land Gini of 0.43, measured around 1995, is from Griffin, Khan and Ickowitz (2002). 28

30 each panel), the FD model exploits exactly two first differences: one taken at the beginning of the price spike and one taken immediately after the shock. Since neither DEMd i nor LGINI i are time-varying, the source of variation in first differences is I HIGH, which is non zero at precisely these two points. The evidence from the FD models is starkly based on difference in differences taken before versus after the price shock. It is therefore sensitive to the timing of the shock. 16 To summarize, our main hypothesis posits that high-price shocks incentivize non-democracies to raise export taxes on products that use land intensively in their production, a factor the elite that are in power own abundantly. By so doing, they attempt to placate their populations and prevent food riots that could potentially escalate and foment revolution. 17 Figures 2 and 3 anticipate the econometric results. The top panel of Figure 2 depicts time series of mean N RA for three groups of countries: non-democracies, middle democracies and liberal democracies (see Appendix Table A1 for countries in the sample and descriptive statistics). Non-democracies, defined as countries with a mean Polity score of less than zero. 18 already had a negative mean NRA in the years immediately prior to the price spikes. Yet, once the spikes hit, their mean N RA dropped sharply. The response of Middle democracies was similar, a result that we investigate further below. Liberal democracies, had a subsidy regime (NRA > 0) to begin with, reduced their subsidies on average, but refrained from taxing their exporters. That is, the effect of markets were muted in non-democracies and middle democracies, on average, while agricultural export prices in liberal democracies remained market determined. The bottom panel of Table 2 depicts mean NRA M over the same period. The response of nondemocracies is clear to see: sharp cuts in assistance (trade protection) to farmers at the onset 16 This distinct difference between the two models in our context is in addition to other reasons for why both FE and FD models should be estimated. FD models provide useful checks on estimates from the FE model if errors are in fact strongly serially dependent. If, for example, the errors e i,t in the FE models followed a random walk, it is more reasonable to assume that the differenced errors are homoskedastic and uncorrelated, making the FD estimates more efficient than FE estimates (Wooldridge, 2002 Ch 10). Also, if the FE and FD estimates differ significantly say they have opposite signs it may indicate that the exogeneity assumptions are violated. The strict exogeneity assumption that makes FE estimates unbiased is the same strict exogeneity assumption that makes FD estimates unbiased, and therefore they should not be very different. If the exogeneity assumption is satisfied, any difference in magnitudes of the estimates is ascribed to the fact that the FD model controls for trends in the error term that the FE model does not capture. We include the initial NRA as a variable in the FD model. The lagged dependent variable is dropped before differencing, on the assumption that the first differenced NRAs are not autocorrelated 17 A similar mechanism works through imports: rather than protect products that use land intensively from import competition, high-price shocks incentivize the elite to cut tariffs and even subsidize imports so to placate their populations. 18 The mean Polity scores to define the three groups of countries in Figure 2 are taken over , and in Figure 3 over

31 of the spikes. Middle democracies also decreased their NRAs on imports but the response was gradual, unlike the non-democracies. Liberal democracies, on the other hand, had started to liberalize since the late 1990s in accordance with cuts agreed at the Uruguay round. Figure 3 presents similar trends even though more non-democracies existed at the time of the 1973 spike than at the time of the 2004 spike. Non-democracies and middle democracies do not appear distinct in their response to the price spike. 6 Results Pre- Versus Post-Price Spike Table 2 presents coefficient estimates from the FE and FD models of nominal rates of assistance to exports NRA from the sample. The first two FE model columns restrict the sample to only non-democracies, defined as countries with a mean Polity2 score (over the sample period) of less than zero. The first column includes the lagged dependent variable, log income and a linear trend as regressors. The pre- versus post-spike difference, measured by the coefficient on It HIGH, is negative and statistically significant in both models. The estimate from the first model implies that, on average, NRA in the price-spike years was 8.8% higher than in the non-spike years. We note that since NRA does not fully rebound in the years immediately following the spike (Figure 2), the estimate understates the extent of the immediate response by non-democratic governments to the onset of the spike. The second column indicates the estimate is robust to including the log of the nominal market exchange rate index (defined as home currency per dollar indexed to 1995 as a base year). OLS-FE estimates from the full sample are presented in the next two columns, where the estimate on I HIGH t continues to be robust. Its interaction with the Middle Democracy indicator and the Liberal Democracy indicator shows whether they, on average, differ in their response to the spikes. The total effects for the two groups are presented in the last two rows. Middle democracies responded no differently, while liberal democracies did not appear to respond at all to the price-spikes. response. The FD models in the four columns on the right of Table 2 show a stronger non-democratic The coefficient on I HIGH t shows non-democracies imposing a tax on farm exports during the spike years that is over 12 percentage points greater than during the non-spike years. The case of Pakistan s export restrictions on rice is telling. Dorosh and Salam (2009) and Salam 30

32 (2009) document the emergence of Pakistan from a regime that severely taxed exports of cotton and rice in the 1970s to a regime that liberalized prices; by 2004, exports of these main products suffered few distortions. The severity of the 2006 price shock threatened to reverse many years of reform. Pakistan s regressed to the old policy of setting low procurement prices for rice and even cotton (inputs into the garments sector that produced export earnings), thus heavily taxing the exports of these sectors. The FD model estimates show that, while middle democracies also raised export taxes during the price spike, it was half the magnitude of the non-democratic response. Liberal democracies did not respond with policy towards agricultural exports during the spike. Table 3 presents estimates from the same models, but examines the non-democratic response to the first post WWII food price spikes in This was a period of sweeping economic and political change. Macroeconomic shocks from the dissolution of the fixed exchange rate regime of Bretton Woods were followed by oil price shocks unleashed by OPEC countries. The price spikes sat on the cusp of waves of democratizations that began in Europe and spread to Latin America (Huntington 1991). The FE models indicates that N RA in non-democracies decreased by approximately 5 percentage points during the price spikes. The OLS-FE results for middle and liberal democracies (from the full sample) are different for the 1970s spike compared with those from the spike. Both groups actually reduced their N RA much more, on average, than did non-democracies. Liberal democracies reduced their N RA by a massive 17 percentage points, and middle democracies by approximately 11 p.p. Liberal democracies, however, were lowering their export subsidies, which had fewer benefits for their populations than the increase in export taxes had on populations in middle and non-democracies by insulating them from the price hikes. Since in many middle democracies a significant proportion of households expended a large proportion of their income on food, the regimes were politically sensitive to food price hikes. The results therefore show the increase in export taxes as a natural response of governments who needed to pay attention median voter concerns. 19 The FD models in the right panel 19 Log income and log exchange rate index are statistically and economically significant for this period. While the positive cross-sectional correlation between income and NRA is well established high-income countries can and do subsidize their farmers (e.g. Anderson xxx) it is a surprising finding in the restricted sample of nondemocracies who generally do not subsidize exports, or after controlling for fixed effects in the full sample. A plausible reason for the finding in the restricted sample is that non-democracies choose between two instruments explicit consumption subsidies versus taxation of exports. In favorable states of the world when GDP increases, those governments choose consumption subsidies (and low export taxes) because they can afford to pay for the subsidies. But in unfavorable states of the world in which negative GDP shocks strap budgets, non-democratic governments choose high export taxes (and low or zero consumption subsidies). The estimates imply that a 10% decrease in per capita GDP increased income taxes (lowered NRA) The negative coefficient on log exchange 31

33 of Table 3 affirm the FE results. Difference-in-Differences: Land Inequality The first test of our theory with the price spike is presented in Table 4. The force of the test is the assumption that in countries with the greatest land inequality, the elite own this precious resource (xxxdocument this from studies!!). Imposing a tax on food exports to reduce domestic consumption prices for the median consumer implies taxing the elite, often severely. This is the trade-off the elite must make to keep political power. The focus is on the difference-in-differences measured by the coefficient on the interaction LGINI I HIGH, where the first difference is taken with respect to the price shock and the second with respect to the land gini. The FE estimates from the abridged sample of non-democracies show this estimate to be large and statistically significant at the 10% level. In the full sample, this estimate for non-democracies is more precisely measured, and robust across the two models. The negative coefficients on the interaction imply that the 2000 s price spike induced a larger export tax in countries with greater land inequality. In the model with the full set of controls, the estimate of implied an export tax that was 7.37 percentage points larger in countries in the third LGINI quartile than in the first LGINI quartile. The estimate implies an export tax differential in the non-democracy with the most unequal landholding (LGINI=0.790) over the least unequal one (LGINI=0.345) of 19.7 percentage points! The FD models in the right panel of Table 4 not only estimate the difference-in-differences with great precision, but, notably, they estimate it to be twice as large as the corresponding FE estimate. We take the estimates as strong affirmation of our model. An unspoken implication of the theory is that this result is special for non-democracies, and should not necessarily hold for democracies. Does this bear out in the results? The total effects at the bottom of the table show this difference-in-differences is less precisely estimated for middle democracies and nonexistent for liberal democracies. The estimate of.290 for middle democracies is still large, at half the magnitude of non-democracies, and hints at a similar political mindset. In that rate indicates that countries that devalued (increased the domestic currency-to-dollar exchange rate) taxed their exports (reduced NRA) more. In (1) a devaluation means a lowering of E, which, all else equal, implies a lower NRA. Even if the world price did not change, a devaluation imposed a tax on the exporter. While log exchange rates did not appear to be a factor in the spike perhaps because many countries had moved to less managed exchange rate regimes devaluations were a feature of exchange rate regimes in the 1970s (Edwards and Santaella 1993). 32

34 sense they are better described as illiberal democracies where weakness in institutions makes plausible the threat of being overthrown by revolution. In liberal democracies, the force of theory is vindicated. Not subject to the threat of revolution means their policy in the face of a food price shock is not driven by land inequality. This is despite the fact that the sample of liberal democracies is characterized by significant land inequality; their interquartile range for LGINI is 0.240, compared with for non-democracies (and for middle democracies). Democracy Measures Is this result special to the particular measure of democracy we have adopted? Polity scores are built around the concept of democracy as resting on three fundamental pillars: Executive Recruitment, Executive Constraints and Political Competition. The idea behind Polity scores is to map these latent concepts into precise quantitative measures. The latent concept Executive Recruitment, which seeks to measure how open and competitive is the process of selection and recruitment of people who will assume executive power, is then measured by the variables XRCOMP and XROPEN, respectively. XRCOMP scores countries on a 3 point scale based on whether executive recruitment is based on elections (score of 2) or selection (score of 0). XROPEN is a binary measure of whether executive is elected (=1) or determined by hereditary selection (=0). The latent concept Executive Constraints seeks to measure checks and balances on the executive powers. The variable XCONST maps this concept to a 7 point scale based on whether the executive is subordinate to a parliament or the judiciary (=7), has greater power than but is constrained by parliament or judiciary (5,6), is subject to a few limited constraints (3,4), is unconstrained or has unlimited authority (1,2). The latent concept Political Competition seeks to measure the degree to which political participation is competitive and open. Two variables, PARCOMP and PARREG are operationalized and combined to form this measure. PARCOMP measures competitiveness on 5 point scale based on whether formation of political parties is totally repressed (=1), restricted (=2), factional or represents an electoral transitional (=3,4) or unfettered (=5). PARREG uses a 5 point scale to measure openness based on whether formation of political parties is restricted or regulated (=4, 5), sectarian or identity-based (=2,3) or unregulated (=1). Using each Polity pillar as the basis for separating our sample into the three democracy 33

35 types, we re-estimate the models. 20 Table 5 reports these results (for brevity only the results on the interactions are presented). The bottom row shows how the sample breaks down by the three regime types, showing ample variation in their composition. The robustness of the main result the negative difference-in-differences estimate for the effect of land inequality on the response of non-democracies to food price shocks is therefore both surprising and reassuring. The affirmation of the theory is robust also to whether this difference-in-differences is estimated from the FE or FD models. models exploit different sources of variation. We feel this is significant since, as we have described, the two The FD model estimates are driven by firstdifferences taken at precisely two points (for each country): at the inception of the shock and at the first period after the shock dissipates. The FE model estimates rely on pooling withinvariation across countries. The land inequality difference-in-differences for non-democracies ranges between and At the lower end the estimate implies that non-democracies in the third land inequality quartile impose an export tax (in response to the food shock) that is, on average, 8 percentage points higher than non-democracies in the first land inequality quartile. At the higher end the difference is 27.4 percentage points! We are properly aware that due to the small sample of non-democracies these statements are perhaps strong. But the robustness across models and democracy measures indicates the resilience of the estimates of the land inequality effects. No such results are robustly evident in middle and liberal democracies, additional affirmation for the theory s special applicability only to regimes where the elite feels threatened with revolution. Another way of saying this is that a form of the revolution constraint applies to every regime the executive can be rejected in liberal and middle democracy at the polls. However, this form of rejection is less threatening to an elected executive than it is for an selected elite being overthrown by a revolution or coup. It is perhaps also generally true that the spoils from ruling over a non-democracy with few constraints or checks on their power selected are far greater than the rewards that accrue to elected executives in liberal democracies, and therefore the incentives to survive shocks that much stronger. The 1970s Shock 20 Countries are partitioned into {non-democracy}, {middle democracy}, {liberal democracy} as follows: PAR- COMP: {1,2}, {3,4}, {5}. POLCOMP: {1,2,3}, {4,5,6,7,8}, {9,10}. XRCOMP: {0,1}, {2}, {3}. XCONST1: {1,2}, {3,4,5,6}, {7}. XCONST2: {1,2,3}, {4,5,6}, {7}. Note that POLCOMP is the latent concept that is measured on a 10-point scale on the basis of its components PARCOMP and PARREG. 34

36 Are our results special to the 2000 s price shock or is this a generally robust finding of about food price shocks? We indicated in Table 3, another significant instance of a massive food price shock, the price spikes. This period also affords us data on many more non-democracies than the 2000s sample. However, we must use the same land inequality measures because those are the only ones available. This requires the assumption that land inequality is a persistent phenomenon, resistant to change. 21 which includes 24 non-democracies. 22 the theory. We begin in Table 6, with the shocks, a period The models provide a surprisingly strong affirmation of Even with a widely different group of non-democratic countries compared with the 2000 s sample, the FE model estimates the land inequality difference in differences to be in the full model. This implies that non-democracies in the top land inequality quartile taxed exports, in the face of the 1970s price shock, by 3.83 percentage points more than nondemocracies in the bottom land inequality quartile, on average. The same difference between the countries with the highest and lowest land inequality was 10.2 percentage points. While the magnitude of this estimate is half that for the 2000 s shock, it is precisely estimated and its affirmation of the theory should not be surprising. Many of these countries became recently independent of their colonists, with nascent institutions and little tax capacity. These should behave just as theory predicts because they come close to fulfilling many of the assumptions underlying the theory. At the same time there are a number of countries that were transitioning to democracy (Portugal, Spain, Argentina, Brazil, Chile). The theory perhaps applies less to them, although we have seen the clear distinction between the liberal democratic response and the middle-democratic response. The embryonic phase of their institutions, where such existed, made their behavior distinctly different from that of more mature liberal democracies. Another result from the FE model that is somewhat noteworthy is the same estimate for liberal democracy (reported at the bottom) is exactly the opposite: countries with high land inequality maintained free trade and even increased export subsidies. 23 This is rather in line with the prediction of the Grossman-Helpman lobbying model in which subsidies were 21 A number of studies show that initiation of land reforms have had little impact on land inequality, where land reforms are rolled back under new governments. Binswanger et al. (1995) cover many regions; Deininger and Squire (1998, Table xx) indicates the stasis. Faquet et al. (200xx see results folder) provides a survey of the South American experience with land reform, particularly Columbia. 22 Arranged in ascending order of average Polity score over the sample period, the non-democracies are: Cote D Ivoire, Cameroon, Chad, Indonesia, Kenya, Mali, Togo, Zambia, Egypt, Philippines, Sudan, Brazil, Senegal, Benin, Argentina, Ghana, Ecuador, Madagascar, Chile, Spain, Dominican Republic, Thailand, Bangladesh, Portugal. 23 While this effect is less precisely estimated in the FD models, the effects are nonetheless economically significant for their magnitude. 35

37 for sale to land-owning agricultural lobbies. This finding clearly distinguishes the two models. The Grossman-Helpman model s prediction is clearly inapplicable to non-democracies, where lobbying institutions are absent or ineffective. And the diametrically opposite finding for liberal democracies from that of non-democracies begs a model founded on non-democratic institutions, such as ours. It is likely that since the spikes posed these countries the first food shock of this magnitude, they reacted with a lag. If we build in a one-period lagged response of the NRA to the shock, the results are stronger in favor of the theory. These are presented in Table The non-democratic response to the food price spike varied significantly with land inequality. The difference across the two land inequality quartiles is a 6.1 percentage point higher export tax, on average, in the full FE model and 7.2 percentage points in the corresponding FD model. We also admit to not finding the same lagged response effect to the more 2000 s shock. Plausibly, this first shock caught the ruling elites of the many newly independent countries unaware, resulting in learning before their policy response. By the 2000s, most countries were seasoned in the art of using trade policy to keep power, and reacted quickly. At least this is how we see it. A VAR study of response to these two shocks would provide more convincing evidence. Finally, in Table 8 we subject the results of Table 7 to robustness of the definition of non-democracy. While the set of non-democracies stays at 24 through the different definitions, and therefore produces similar results to table 7, with XCONST1 the non-democracy sample is reduced to 18 and the results more or less holds. In unreported results, we further tighten each definition so the number of non-democracies is reduced (and their number varies across definitions), and the result remains robust. 25 The number of liberal and middle democracies varies significantly across the columns, producing swings in their estimates of the land inequality effect. They do not prove robust to different definitions of democracy. 7 Conclusion The purpose of this paper has been to explore, theoretically and empirically, how dictatorships use international trade policy to forestall democratization. The framework makes it possible 24 In this table, we allow the shock to end in 1974, that is, it is a two-period shock. The FD models are especially sensitive to the timing of the shock. 25 These results are available from the authors. 36

38 to consider the role of world food price shocks in triggering a threat of revolution. This is a new contribution to the literature, in that prior research in economics on democratization has tended to assume that the threat of revolution arises exogenously. The model predicts that when a dictatorship has a comparative advantage in primary products and the elite are land owners, they have an incentive to forestall democratization by increasing export taxes. On the other hand, when the country in question has a comparative dis-advantage in primary products (while the elite are land owners) the elite have an incentive to forestall democratization by raising import tariffs. Using data from , we found supportive evidence of these predictions for the food price shocks of and An important aspect of our analysis is that it provides predictions for dictatorships that contrast with those of GH, which is the leading model of trade policy for established liberal democracies. More significant still, the theoretical predictions that we derive for dictatorships also seem to predict the trade policy responses to world food price shocks of what we call middle democracies as well. This suggests that middle democracies may have more in common with dictatorships than liberal democracies. In particular, it may be that at least some of the middle democracies in our dataset can better be characterized as so-called illiberal democracies. That is, although they share some superficial features with democracies such as elections, in fact power is more likely to change hands through political violence than at the ballot box. Consequently, political leaders in illiberal democracies may have an incentive to use trade policy in much the same way as dictators do in response to world price shocks. Future research: When do dictators fail in their efforts to forestall democratization using international trade policy? That is, can we identify situations where world food price shocks have caused a transition to democracy? Could a structural econometrics approach shed light on counter-factuals? What happens to illiberal democracies? Do they typically evolve towards democracy or lapse back towards dictatorship? 37

39 References xxxxland inequality cites, incl sources I used to fill in missings xxx Abderrahim, Kaouther and Vincent Castel Inflation in Tunisia: Perception and Reality in a Context of Transition. African Development Bank (Chief Economist Complex) Anderson, K Lobbying Incentives and the Pattern of Protection in Rich and Poor Countries. Economic Development and Cultural Change 43(2): Anderson, K., M. Kurzweil, W. Martin, D. Sandri and E. Valenzuela Measuring Distortions to Agricultural Incentives, Revisited. World Trade Review 7(4): Anderson, K., ed Distortions to Agricultural Incentives, A Global Perspective, 1955 to Washington, DC: The World Bank and Palgrave Macmillan Anderson, K., and E. Valenzuela Estimates of Distortions to Agricultural Incentives, 1955 to 2007, core database at Bates, R.H. 2007a, Domestic Interests and Control Regimes, p in B. Ndulu, P. Collier, R. Bates and S. O Connell (eds.) The Political Economy of Economic Growth in Africa, Cambridge: Cambridge University Press. Bates, R. H. 2007b. Political Reform, p in B. Ndulu, P. Collier, R. H. Bates and S. O Connell (eds.). The Political Economy of African Economic Growth, Cambridge: Cambridge University Press. Bates, R. H Markets and States in Tropical Africa. Berkeley and Los Angeles: University of California Press Bazzi, S. and Blattman, C Economic shocks and conflict: evidence from commodity prices. American Economic Journal: Macroeconomics 6: 1E38. Berazneva, J. and Lee, D Explaining the African food riots of 2007E2008: an empirical analysis. Food Policy 39: 28E39. Biello David Are high food prices fueling revolution in Egypt? Scientific American 1/ Binswanger, Hans. P., K. Deininger, and G. Feder Power, Distortions, Revolt and Reform in Agricultural Land Relations. In J. Behrman and T. N. Srinivasan (eds.) Handbook of Development Economics, Vol. 3. Amsterdam: Elsevier. Bruckner, M. and Ciccone, A International commodity prices, growth and the outbreak of civil war in sub-saharan Africa. Economic Journal 120: 519E34 38

40 Dorosh Paul A., Abdul Salam Distortions to Agricultural Incentives in Pakistan.. In K. Anderson and W. Martin (eds.) it Distortions to Agricultural Incentives in Asia. Washington DC: World Bank. Edwards, Sebastian and Julio Santaella Devaluation Controversies in the Developing Countries: Lessons from the Bretton Woods Era. In Bordo, Michael D. and Barry Eichengreen (es.) A Retrospective on the Bretton Woods System: Lessons for International Monetary Reform. Chicago, IL: University of Chicago Press. de Ferranti, David, Guillermo E. Perry, Francisco H. G. Ferreira, and Michael Walton Inequality in Latin America: Breaking with History? Washington, DC. The World Bank. Frankema, Ewout The colonial roots of land inequality: geography, factor endowments or institutions? Economic History Review: Goldstein, Markus and Christopher Udry The Profits of Power: Land Rights and Agricultural Investment in Ghana. Journal of Political Economy 116(6): Griffin, Keith, Azizur Rahman Khan, and Amy Ickowith Poverty and the Distribution of Land. Journal of Agrarian Change 2(3): 279E30. Huntington, Samuel P The Third Wave: Democratization in the Late Twentieth Century. Norman, OK: University of Oklahoma Press. Ianchovichina, Elena, Josef Loening and Christina Wood How Vulnerable are Arab Countries to Global Food Price Shocks? Journal Of Development Studies 50(9). Marshall Monty G, Gurr Ted R., Davenport Christian, Jaggers Keith Polity IV Project: Comparative Political Studies 35(1): 40E45. Marshall Monty G, Gurr Ted R, Jaggers Keith Polity IV Project: Political Regime Characteristics and Transitions, 1800E2012. Dataset userse manual Mayfield, James. B Agricultural Cooperatives: Continuity and Change in Rural Egypt. In Shimon Shamir (ed.) Egypt From Monarchy to Republic: A Reassessment of Revolution and Change. Boulder: Westview Press. Moyo, Sam The Politics of Land Distribution and Race Relations in Southern Africa. United Nations Research Institute for Social Development Paper No: 10. Salam, Abdul Distortions in Incentives to Production of Major Crops in Pakistan: Journal of International Agricultural Trade and Development 5(2): Sumner, D. A Recent Commodity Price Movements in Historical Perspective. American Journal of Agricultural Economics 91: van Weezel, Stijn Food imports, international prices, and violence in Africa. 39

41 Oxford Economic Papers doi: /oep/gpw015. Vollrath, Dietrich Land Distribution and International Agricultural Productivity. American Journal of Agricultural Economics 89(1): Wooldridge, Jeffrey M Econometric Analysis of Cross Section and Panel Data, 2nd edition. Cambridge, MA: MIT Press. 40

42 References [1] Acemoglu, D., S. Johnson, and J. A. Robinson, (2001); The Colonial Origins of Comparative Development: An Empirical Investigation. American Economic Review, 91(5): [2] Acemoglu, D., S. Johnson, and J. A. Robinson, (2005); Institutions as a Fundamental Cause of Long-Run Growth. Published in Handbook of Economic Growth, ed. Philippe Aghion and Steven Durlauf, Amsterdam: North Holland. [3] Acemoglu, D., and J.A. Robinson, (2000); Why Did the West Extend the Franchise? Democracy, Inequality and Growth in Historical Perspective. Quarterly Journal of Economics, 115: [4] Acemoglu, D., and J.A. Robinson, (2001); A Theory of Political Transitions. American Economic Review, 91(4): [5] Acemoglu, D., and J.A. Robinson, (2006); Economic Origins of Dictatorship and Democracy, Cambridge University Press, New York, NY. [6] Aidt, T.S., and M. Gassebner, (2010); Do Autocratic States Trade Less? World Bank Economic Review, 24(1): [7] Aidt, T.S., and P.S. Jensen, (2009); Tax Structure, Size of Government and the Extension of the Voting Franchise in Western Europe, International Tax and Public Finance, 16: [8] Aidt, T.S., and P.S. Jensen, (2014); Workers of the World, Unite! Franchise Extension and the Threat of Revolution in Europe, European Economic Review 72: [9] Anderson, K., and E. Valenzuela, (2008); Estimates of Global Distortions to Agricultural Incentives, 1955 to World Bank, Washington D.C. (Available at website). [10] Bagwell, K., and R. Staiger, (1990); A Theory of Managed Trade. American Economic Review, 80, [11] Bagwell, K., and R. Staiger, (1999); An Economic Theory of the GATT. American Economic Review, 89(1): [12] Bagwell, K. and R. Staiger, (2003); Protection and the Business Cycle. Advances in Economic Analysis & Policy, Berkeley Electronic Press, 3(1), Article 3. [13] Besley, T. and R. Jarayaman, (2010); Introduction. Published in T. Besley and R. Jarayaman (eds.) Institutional Microeconomics of Development, MIT Press, Cambridge, Massachusetts, and London, UK. [14] Besley, T., and T. Persson, (2009); The Origins of State Capacity: Property Rights, Taxation, and Politics. American Economic Review, 99(4): [15] Besley, T., and T. Persson, (2011); Pillars of Prosperity: The Political Economics of Development Clusters, Princeton University Press, Princeton NJ. 41

43 [16] Burke, P.J., and A. Leigh (2010); Do Output Contractions Trigger Democratic Change? American Economic Journal: Macroeconomics, 2(4): [17] Chipman, J., (1969); Factor Price Equalization and the Stolper-Samuelson Theorem. International Economic Review, 10: [18] Chor, D., (2010); Unpacking Sources of Comparative Advantage: A Quantitative Approach. Journal of International Economics, 82(2): [19] Costinot, A., (2009); On the Origins of Comparative Advantage. Journal of International Economics, 77: [20] Decker, J., and J. Lim, (2009); Democracy and Trade: An Empirical Study. Economics of Governance, 10(2): [21] Do, Q.-T., A.A. Levchenko, and M. Ravallion, (2014); Trade Insulation as Social Protection. Published in J.-P. Chavas, D. Hummels, and B. Wright (eds) The Economics of Food Price Volatility, NBER and University of Chicago Press. Also World Bank Policy Research Working Paper [22] Engerman, S.L., and K.L. Sokoloff, (1997); Factor Endowments, Institutions, and Differential Paths of Growth Among New World Economies: A View from Economic Historians of the United States. Published in S. Harber How Latin America Fell Behind, Stanford University Press, Stanford. [23] Farrell, S., (2010); A First Step Towards Democracy. History Today, 60: [24] Galiani, S., and G. Torrens, (2014); Autocracy, Democracy and Trade Policy. Journal of International Economics, 93(1): [25] Garfinkel, M., S. Skaperdas, and C. Syropoulos, (2008); Globalization and Domestic Conflict. Journal of International Economics, 76(2): [26] Garrett, G., (2000); The Causes of Globalization. Comparative Political Studies 33(6 7): [27] Goldberg, P.K., and G. Maggi, (1999); Protection for Sale: An Empirical Investigation. American Economic Review, 89(5): [28] Grossman, G., and E. Helpman, (1994): Protection for Sale. American Economic Review 84(4): [29] Hillman, A.L., (1982); Declining Industries and Political-Support Protectionist Motives. American Economic Review, 72(5): [30] Hillman, A.L., and H.W. Ursprung, (1988); Domestic Politics, Foreign Interests, and International Trade Policy. American Economic Review, 78(4): [31] Jack, W., and R. Lagunoff, (2006); Dynamic Enfranchisement. Journal of Public Economics, 90(4-5): [32] Jones, R.W., (1965); The Structure of Simple General Equilibrium Models. Journal of Political Economy, 73(6):

44 [33] Kindleberger, C.P., (1975); The Rise of Free Trade in Western Europe, Journal of Economic History, 35(1): [34] Krusell, P., and J.-V. Rios-Rull, (1996); Vested Interests in a Positive Theory of Stagnation and Growth. Review of Economic Studies, 63(2): [35] La Porta, R., F. Lopez-de-Silanes, A. Shleifer and R.W. Vishny, (1997); Legal Determinants of External Finance. Journal of Finance, 52(3): [36] Levchenko, A.A., (2007); Institutional Quality and International Trade. Review of Economic Studies, 74:3 (July 2007), [37] Levchenko, A.A., (2013); International Trade and Institutional Change. Journal of Law, Economics, and Organization, 29(5): [38] Liu, X., and E. Ornelas, (2014); Free Trade Agreements and the Consolidation of Democracy. American Economic Journal: Macroeconomics, 6(2): [39] Lizzeri, A., and N. Persico, (2004); Why Did the Elites Extend the Suffrage? Democracy and the Scope of Government, with an Application to Britain s Age of Reform. Quarterly Journal of Economics, 119(2): [40] Llavador, H., and R.J. Oxoby (2005); Partisan Competition, Growth, and the Franchise. Quarterly Journal of Economics, 120(3): [41] Magee, S.P., W.A. Brock, and L. Young, (1989); Black Hole Tariffs and Endogenous Policy Theory: Political Economy in General Equilibrium, Cambridge University Press, Cambridge, UK. [42] Maggi, G., and A. Rodríguez-Clare, (1998); The Value of Trade Agreements in the Presence of Political Pressures. Journal of Political Economy, 106(3): [43] Mayer, W., (1984); Endogenous Tariff Formation. American Economic Review, 74(5): [44] McCormick, John P., (2012); Niccolò Maciavelli ( ). University of Chicago Department of Political Science Working Paper Series.h [45] North, D., (1981); Structure and Change in Economic History, Norton, New York. [46] Nunn, N., (2007); Relationship-Specificity, Incomplete Contracts, and the Pattern of Trade, Quarterly Journal of Economics, 122(2): [47] Olson, M., (1982); The Rise and Decline of Nations: Economic Growth, Stagflation, and Economic Rigidities. New Haven, CT: Yale University Press. [48] O Rourke, K.H., and A.M. Taylor, (2007); Democracy and Protectionism. Published in T.J. Hatton, K.H. O Rourke and A.M. Taylor (eds) The New Comparative Economic History: Essays in Honor of Jeffrey G. Williamson, MIT Press, [49] O Rourke, K.H., and J.G. Williamson, (1999); Globalization and History: The Evolution of a Nineteenth-Century Atlantic Economy. MIT Press, Cambridge, Massachusetts, and London, UK. 43

45 [50] Rodrik, D., (2007); One Economics, Many Recipes: Globalization, Institutions, and Economic Growth. Princeton University Press, Princeton NJ. [51] Rogowski, R., (1989); Commerce and Coalitions: How Trade Affects Domestic Political Alignments. Princeton University Press, Princeton NJ. [52] Schonhardt-Bailey, C., (2006); From the Corn Laws to Free Trade: Interests, Ideas, and Institutions in Historical Perspective. MIT Press, Cambridge, Massachusetts, and London, UK. [53] Sharma, R., (2011); Food Export Restrictions: Review of the Experience and Considerations for Disciplining Restrictive Measures. FAO Commodity and Trade Policy Research Working Paper 32, Food and Agriculture Organization. [54] Smith, A., (1776); An Enquiry into the Nature and Causes of the Wealth of Nations. Republished in 1977 by University of Chicago Press. [55] Solar, P.M., (1997); The Potato Famine in Europe. Published in C. O Grada (ed. 1997) Famine 150: Commemorative Lecture Series. University College, Dublin. [56] Williamson, J.G., (2012); Commodity Prices over Two Centuries: Trends, Volatility and Impact. Annual Review of Resource Economics 4: [57] World Bank, (2008); Rising Food Prices: Policy Options and World Bank Response. World Bank, Washington D.C. [58] Zakaria, Fareed, (1997); The Rise of Illiberal Democracy. Foreign Affairs, 76: [59] Zissimos, B., (2014); A Theory of Trade Policy under Dictatorship and Democratization. Discussion Papers 1403, Exeter University, Department of Economics. [60] Zissimos, B., (2017); A Theory of Trade Policy under Dictatorship and Democratization. Journal of International Economics 109:

46 45

47 Figure 2: Price Spike and the NRA X and NRA M Response. 1

48 Figure 3: Price Spike and the NRA X and NRA M Response. 1