Cursed Resources? Political Conditions and Oil Market Outcomes

We analyze how a country's political institutions affect oil production within its borders. We find a pronounced negative relationship between political openness and volatility in oil production, with democratic regimes exhibiting less volatility than more autocratic regimes. This relationship holds across a number of robustness checks including using different measures of political conditions, instrumenting for political conditions and using several measures of production volatility. Political openness also affects other oil market outcomes, including total production as a share of reserves. Our findings have implications both for interpreting the role of institutions in explaining differences in macroeconomic development and for understanding world oil markets.


I. Introduction
Development levels vary considerably across countries, and explaining why these differences exist is one of the central questions in economics. While scholars have long hypothesized that political institutions play an important role (see, e.g., North 1981), a recent literature has made progress identifying a causal impact of institutions on cross-country differences in macroeconomic outcomes. 1 Still, the mechanisms by which specific institutions affect economic outcomes remain a "'black box'" Robinson, 2001hereafter AJR -p.1395). 2 This paper focuses on the influence of institutions on one particular industry: crude oil production. We document a pronounced negative effect of good political conditions on volatility in oil production, and this result is robust to using several different measures of political conditions. We also address the potential endogeneity of political structure, as suggested by the literature on the "resource curse." We estimate two-stage least-squares regressions where we instrument for recent political institutions with both institutions before oil was commercialized and, following AJR, settler mortality. We find that the negative relationship persists in these specifications. 3 We also evaluate whether other macroeconomic differences across countries affect oil market outcomes, including financial openness and legal formalism. None of these factors has a significant effect on volatility once we control for institutions.
It is important to acknowledge that a country's oil production may not necessarily be inefficient if it is volatile. To allow for this, we also construct volatility measures that control for market factors which should drive changes in a country's oil output, and our results are robust to using these alternative measures. 4 Ideally, we would also like to measure oil output controlling for the natural resources available to the country, as this is analogous to studies that examine factors contributing to differences in economic development. With this goal in mind, we also examine the link between political institutions and a country's average production as a share of its proven reserves. We find that countries with better political institutions produce a higher share, although we devote less attention to these results since the data on reserves are self-reported in some cases and could be systematically biased. Finally, we show that political institutions lead to volatility in the number of active wells, suggesting that the volatility is less likely to be driven by physical characteristics of a country's oil fields.
Oil production is a particularly convenient lens with which to view the possible microeconomic foundations for macroeconomic outcomes as it is a commodity sold on a world market. This means that production decisions within a country should be driven by world demand and not local macroeconomic conditions. Was this not the case, our results would be less interesting as they could simply reflect the effect of political institutions on overall macroeconomic activity. Since oil demand is worldwide, however, we can be confident that our results reflect supply-side factors and not demand-driven output volatility.
Another reason that oil is well-suited to this analysis is that the unit of output (a barrel of oil) is of essentially homogenous quality and is consistently measured across countries. Finally, rich data are available on oil production and its determinants, such as reserves and the number of wells.
By documenting that political institutions affect outcomes in a particular industry, our results provide insight on how institutions affect aggregate economic output. For example, they suggest that while political institutions may work by influencing structural factors, such as by promoting a well-developed commercial sector or less reliance on agriculture (see, e.g., Duarte and Restuccia, 2010), these cannot be the only mechanisms at play as institutions have effects within an industry. Although our results cannot address this hypothesis directly, they are consistent with the idea that well-functioning political institutions support efficient investment in crude oil production.
Our results also provide insight on world oil markets. Like many energy markets, oil markets are characterized by extremely inelastic short-run supply and demand, meaning that even small fluctuations in either can lead to large swings in price. While much has been written about how systematic shifts in oil demand or supply affect prices (see, for example, Hamilton and Kilian (2009)), less is understood about the underlying determinants of short-run changes in supply.
Our results imply that political institutions in the countries endowed with oil reserves affect the volatility, and perhaps the level, of its production. To the extent countries with poor political conditions will have less stable production of other natural resources, our results suggest a useful metric for comparing different energy sources. As U.S. policymakers attempt to drive shifts to new, alternative energy technologies, it is useful to be able to make these comparisons.
To explore the implications of our results for world oil markets, we construct annual, worldwide statistics that summarize the political institutions of oil-producing regimes.
Generally, our indexes depict a reduction in the political conditions of oil producing countries between 1965 and 1978, followed by an increase that peaks sometime in the late 1990s or early 2000s, depending on which index is considered. In recent years most of the indexes show a modest decline starting around 2003.
We next decompose the measures into "production share" and "internal conditions" indexes to show how changes in the aggregate measures are driven by changes across countries in their share of world oil production and changes within countries in political conditions. In the early part of our sample, changes in oil political conditions were mainly driven by changes in the production share, as Middle Eastern countries accounted for a larger and larger share of total production. Recent trends, however, appear to be driven by changes in internal conditions in addition to changes in the location of oil production.
We show that the higher the share of oil coming from countries with poor political conditions in any given year, as measured by the decomposed "production index," the higher is the volatility of oil prices in that year. Within-country changes in political conditions, as measured by the decomposed "internal conditions index," if anything have the opposite effect on oil price volatility, suggesting that it may take time before short-run changes in political conditions impact oil production.
While short-run production volatility within one country's borders may be quickly counteracted by adjustments in other countries so that any resulting price volatility will be shortlived, there are reasons to be concerned about the volatility. For one, price volatility makes business planning difficult and raises the cost of hedging risk. It is particularly difficult for energy-dependent industries such as the airlines and automakers. It also raises concerns for many that speculation is contributing to price volatility and spikes causing many to call for tighter regulation of energy markets. 5 The next section of the paper presents a model that helps elucidate the mechanisms by which political conditions might affect oil markets. Section III presents our main empirical findings, documenting the negative relationship between political conditions and oil market volatility. Section IV interprets our findings and analyzes their implications for price volatility in world oil markets. We also discuss why our findings probably do not support an "energysecurity" based argument for government intervention in energy markets. We conclude in section V with some thoughts on the lessons to be drawn from this analysis as well as possible future extensions of this research agenda.

II. Background and Theory
In this paper we seek to identify the role that political conditions play in driving the volatility of oil production. While a considerable literature exists on the causal relationship between a country's oil wealth, or wealth from other natural resources, and political governance (the "resource curse" literature), there has been little focus on the role governance plays in affecting supply. We discuss the resource curse literature in this section both to distinguish our paper from that line of research and to motivate the econometric strategy we use in the next section.
A number of papers have analyzed the relationship between a country's resource wealth and various economic and political outcomes, including slower economic growth and development, conflict, and corruption. We focus on the subset of papers that argue that "oil and mineral wealth tends to make states less democratic" (Ross (2001), p. 328). Ross (2001) identifies three causal mechanisms: governments keep tax rates low and spending high to dampen support for democracy (a rentier effect); they spend money on police to repress dissent (a repression effect); and they fail to develop a modern economy with industrial and service jobs thereby preventing the rise of a middle class that might demand greater voice in government (a modernization effect). Haber and Menaldo (2008) find support for a resource curse of the sort articulated by Ross in simple pooled time-series cross-section regressions but they find the results are sensitive to the inclusion of fixed effects. See also the recent empirical analysis by 5 The Commodity Futures Trading Commission announced plans to consider placing limits on the trading behavior of financial investors in futures markets for energy products. The CFTC opened hearings in July 2009 to consider how this might be best accomplished. See Andrews (2009). Wacziarg (2009). For our purposes, these three papers (and others such as Smith (2004), Basedau and Lacher (2006), Tsui (2008), and Alexeev and Conrad (2009)) highlight the point that we cannot treat political institutions as exogenous in any analysis of the role political conditions play in affecting oil supply, and they motivate our choice of instruments for a country's political conditions.
Turning to the affect of institutions on energy markets, Bohn and Deacon (2000) consider how ownership risk affects the exploitation of natural resources in countries. They find that higher ownership risk decreases oil drilling and oil production whereas higher risk increases deforestation. Ownership risk affects oil and forestry differently due to differential capital intensity. While ownership risk encourages more rapid exploitation to extract resources before assets are (potentially) expropriated by the government, capital investment required to extract resources is also discouraged. Since oil production is relatively capital intensive (compared to forestry) this investment effect offsets the production effect. Our paper differs from Bohn and Deacon in several important respects. First, while they interpret weak institutions as indicative of "ownership risk," in one specification, we include a variable that measures expropriation events in addition to the political conditions variable. The results of this specification suggest that political conditions affect oil production more through other channels (e.g., government pressures to pursue nonmarket goals) than through expropriation risk, although this is based on only one, perhaps blunt, measure of ownership risk. Second, we treat governance institutions as endogenous and instrument for them in our empirical analysis.

A. Model Setup
We consider a simple model that illustrates the relationship between political structure and production volatility. We take as our point of departure the model of Acemoglu and Robinson (2008) which describes the relationship between political and economic institutions.
There are two political groups, an elite (E) with M E members and citizens (C) with M C >M E members. Members of each group wish to maximize the present discounted value of consumption over an infinite horizon, ∑ t t t c β , where β is a discount factor common to all and c t is consumption for an individual in period t.
Each period the groups struggle for control of the government. The group in power controls economic resources in a country and wishes to maximize economic rents for itself. In particular, the government controls a nationalized oil sector and receives the revenue from oil sales. 6 We model political institutions as a state variable in a Markov model in which economic decisions along with citizen mobilization affect the probability of future political institutions.
Thus there is an interplay between political institutions and economic outcomes.
A state variable s indicates whether the political regime is a democracy (state D) or autocracy (state A). Which state the regime is in depends on the outcome of a struggle between citizens and the elite as described below.
As discussed below all decisions made in a given time period affect only that period with the exception of the value of the political state variable which sets the regime type for the following period. Thus, in the discussion below we ignore time subscripts as all decisions relate to the current period (except where specifically noted otherwise).

Oil Production and Revenue Distribution
The government controls oil production in a country and must decide investment levels for production infrastructure (broadly defined as described below) and accrue rents for itself.
For simplicity assume that oil production is the only source of revenue for the government. Each period the group controlling the government invests K in the sector financed by a lump-sum tax paid by all residents of the country. This includes spending on new capital and maintenance of existing capital as well as training of workers and managers. It could also include spending to hire higher quality managers. Higher levels of K reduce the probability of oil production disruptions. 7 Capital is fully depreciated in one period of the model.
An autocratic regime may also have a greater ability to monitor and stifle political dissent that could take the form of activities that reduce oil supply (Ross's Repression Effect). This could include sabotage of capital infrastructure as well as labor slowdowns among the workforce. 8 Alternatively, democratic governments provide additional outlets for dissent that mean workers need not sabotage energy capital to be heard by political leaders. 6 The model could be extended to allow for a capitalist economic system with corporate ownership of natural resources. In that case, tax and royalty policy would be potential instruments to extract oil rents. 7 We ignore the possibility of foreign investment in the domestic oil sector. This would provide another channel of potential oil production volatility. Weak property rights that may be more likely in a less open political environment will deter investment. 8 Nigerian rebels have disrupted oil production as noted elsewhere in our paper. Venezuelan oil workers went on strike in protest of decisions taken by Hugo Chavez in 2002.
Finally production disruptions can occur unexpectedly and for reasons unrelated to fiscal decisions made by the government. Specifically we assume that the value of oil production (R) in a period is given by the following: and y * is a latent variable defined as In equation (2) δ is a measure of an autocratic government's ability to monitor and deter activities that could disrupt production when s = A (I A is an indicator variable equal to 1 when s = A and zero otherwise). We allow δ to be negative given the possibility that disruptive activities under autocracy more than offset the government's ability to monitor and deter those activities. As we shall see below, the sign and magnitude of δ have no bearing on the degree of production volatility. Finally ε is a random variable drawn from a density function f(ε ) and cumulative distribution function F(ε ). We assume that P(R=R H ) = P( 0 * ≤ y ) is high so that increased volatility is associated with higher values of y * . Higher values of K then are associated with lower production volatility.

Political Struggle
The two groups engage in a political struggle for control of the government. We assume that a more democratic political structure can make it more possible to engage in public organizing to preserve democratic institutions. 9 In addition, the ability to mobilize and coalesce opposition to a regime and force an opening of political institutions has a random component (see Acemoglu and Robinson (2006) for a full discussion of the forces driving regime transition).
Activities that lead to potential change in political institutions occur at the end of the current period and determine political institutions for the following period. 10 The following equations define the political transition process: where s 1 is the political state in the following period and (4) The parameter π measures the greater ability of citizens to mobilize for more democratic institutions if the current regime is democratic (I D equals one when the political state is D and zero otherwise). The random variable η captures the stochastic nature of the public's ability to coalesce its opposition to the elite and is drawn from the density function g(η ) with cumulative distribution function G(η ). We assume that ε and η are independent random variables (though this could be relaxed). 11 As in Acemoglu and Robinson (2008) the winners of the political struggle choose the political regime for the following period. The elite will choose an autocracy since this increases the likelihood of their obtaining oil production revenues while citizens will choose democracy.

Timing of Events
Events occur in the following sequence.
is given exogenously based on actions taken last period.
• Knowing s, the current government chooses K, the investment in the oil sector for the current period (first stage decision). • A draw is taken from the density function f and oil revenues are realized.
• A draw is taken from the density function g. Based on the draw, either the elite or citizens take control of the government. They choose the political regime for the following period. Oil revenues accrue to the political class now in control of government. 10 We have also considered a model in which citizens and members of the elite can invest in de facto power that affects the probability of regime change. The model is considerably more complicated but does not provide additional insight. 11 We have also constructed a model in which economic agents may invest in de facto power to increase the probability that their group will win the political struggle. As in Acemoglu and Robinson (2008), it is never optimal for citizens to invest in de facto power. The resulting solution is more complicated and does not provide clear results on conditions for higher or lower production volatility under either regime.

B. Solving the Model
Recall that the decision over the investment level is made by the group in power at the beginning of the period. Define V E (s) as the value function for a representative member of the elite when the political state equals . Similarly V C (s) is the value function for a representative citizen given state s. Thus a representative member of the elite chooses K in a non-democracy to maximize: , v = H, L and is the difference in the value function for a representative member of the elite when the political state shifts from democracy to autocracy . The function ψ measures the expected net benefit to a member of the elite of oil revenue conditional on oil production. Ex post benefits will depend on the outcome after the political struggle and oil investment is undertaken. First order conditions for K(A) (i.e., K in the autocratic state) are given by (6) holds as an equality if K(A) > 0. We will assume positive values for K in the analysis below. Second order conditions require Assuming an interior solution Investment in oil capital rises with the expected net gain from oil production rising from R L to In democracy a representative citizen chooses the level of oil investment to maximize: is the expected net benefit to a representative citizen of oil production conditional on R v production levels and , the expected net gain in utility to a representative citizen from shifting from a non-democracy to democracy.
First order conditions are Second order conditions are Assuming an interior solution, The relation between production volatility and political structure is a complex one. We first note that an improvement in the ability to monitor and deter oil production disruption in a non-democratic regime ( ) 0 > Δδ has no effect on the equilibrium level of production volatility.
To see this note that the right hand side of equation (8) is unaffected by a change in δ. Thus δ α is unaffected by a change in δ. Increases in δ are offset by reductions in K(A) leaving production volatility unchanged.
Oil production is more volatile as the probability of disruption rises which in turn is related to the amount of capital investment. The model suggests several channels through which the political state can affect investment (and possibly production volatility). They include 1) regime specific ability to monitor and deter production disruption activities (δ) as discussed above; 2) the extent of oil revenue losses due to disruption ( R Δ ); and 3) citizen organizing gains when political state equals D that affect optimal oil revenue sharing (π ).
Volatility is higher in an autocratic political regime (s = A) if the following is true: Assuming the second order conditions for investment hold as a strict inequality, this will occur when (14) Equation (14) can be re-written as In other words production volatility under autocratic governments is more volatile if democracy is sufficiently more persistent than non-democracies. This condition is consistent with the empirical findings of Perrson and Tabellini (2009) that democratic institutions contribute to a consolidation of democratic institutions across countries.
Our model provides some insight into the forces affecting oil production volatility and provides a theoretical link between political structure and economic outcomes in this industry.
We turn next to an analysis of oil production data and political structure as an initial step to developing a deeper understanding of this relationship. In the analysis below, we test empirically whether in fact oil production volatility is higher in non-democracies than democracies.

III. Empirical Analysis
We analyze the empirical relationship between a country's political conditions and its oil production in several steps, first presenting our basic results on political conditions and volatility, then describing our approach to dealing with the possible endogeneity of political conditions, next testing the empirical relationship using several additional political conditions measures, and finally presenting several alternative specifications that evaluate whether other macro factors influence volatility in addition to political institutions. We also show results suggesting that political conditions influence the share of a country's reserves that are extracted, although we place less emphasis on these results since reserve estimates could be misleading. Table 1 provides summary statistics on all of the variables used in the analysis.
Our basic regression is of the form: V is a measure of oil volatility in country i, PC i is a measure of political conditions, X i is a vector of control variables included in some specifications and i η is a random error. We estimate equation (16) across the 48 major oil-producing countries. All of our specifications are weighted by a country's average production within the appropriate time period, although the basic results are not sensitive to this weighting.
Our base measure of political conditions is the Composite Democracy Index from the 2007 Polity IV Project. 12 The composite score is defined as the difference between the Institutionalized Democracy index (DEMOC it ) and the Institutionalized Autocracy index (AUTOC it ). The former index is based on the competitiveness and openness of executive recruitment, constraints on the chief executive, and competitiveness of political participation.
The index ranges from zero to ten. The autocracy index is based on other measures of competitiveness and openness of executive recruitment, constraints on the chief executive, and competitiveness and regulation of political participation. It also ranges from zero to ten. The resulting combined polity score ranges from -10 (strongly autocratic) to +10 strongly democratic.
For use in our index decompositions in Section IV, we rescale the index to range from 1 to 21 with higher values still indicating stronger democratic tendencies. Country-level oil production data is from BP (2009).

A. Basic Results
Table 2 presents versions of equation (16)  institutions has begun with the 1970s to ensure that the countries are independent nation states throughout the entire period of analysis (Acemoglu, Johnson, Robinson and Thaicharoen, 2003).
The results in the first row suggest that countries with higher polity scores have lower volatility, and the coefficient estimate is similar across the two different time periods.
Considering the specification in the left-hand column, the magnitude of the coefficient on polity suggests that moving a country from the 25 th percentile polity score of 4.1 (e.g., Libya or Vietnam) to the 75 th percentile polity of 16.4 (e.g., Thailand or Ecuador), should reduce production volatility by .084, which represents approximately half a standard-deviation change in volatility. Also, the R 2 in that specification suggests that political conditions explain nearly one fifth of the volatility differences across countries. Figure 1 plots the data used to generate this set of results, where the size of each circle is proportional to the country's average production over the sample period.
We also measure volatility in oil production after controlling for changes in underlying market conditions. For example, if a country's oil production increases dramatically over the span of several years because of the discovery of new oil reserves, this would be considered appropriate, economically-motivated volatility. Similarly, profit-maximizing firms within a democratic country may cut back on production if world demand declines.
To separate the "excess" volatility from the market-driven changes in oil production, we consider the following formula: and then calculate the residual production: Our country-level measure of volatility is then: The second row of Table 2 reflects a specification where f() is a country-specific quadratic time trend. At the country level, scientists have suggested that production will increase slowly at first, then more quickly and eventually decline (Hubbert (1956)). Because some countries began extracting oil long before the beginning of our dataset while others began during our dataset, we are concerned that the volatility measure might reflect changes in output driven by petroleum engineering factors, as countries with recent oil discoveries would go through both the slow and rapid increase in our data set. The results in the second row suggest that controlling for this possibility, the relationship between polity and volatility is even stronger, and, in the 1981-2007 time period accounts for almost half of the variation in volatility. 13 The third and fourth rows of Table 2 estimate f() using information on world GDP, which affects world oil demand, and a country's reserves, which, as an abstraction, we take to be exogenous determinants of the country's ability to produce oil. 14 The reported specifications reflect a coefficient of one on both ln(World GDP) and ln(Reserves), although the results are quantitatively very similar if we impose different coefficients on either variable. 15 In both rows, the absolute value of the coefficient on polity declines, but remains statistically significant at the five-percent level.
Finally, the last row of Table 2 reports coefficient estimates that use a dependent variable measuring the largest single percent decline in annual production. The results similarly suggest that countries with poor political conditions are more prone to large output drops. In the regressions that follow, we will report results based on the volatility measure used in the first row of Table 2, although our results are not sensitive to this decision.

B. Reverse Causality
We argued in section II that our polity measure is likely endogenous and possibly correlated with the error term in regression equation (16). In Table 3 we report results using several different instruments, all of which are based on the assumption that activities which took place before oil was commercialized in a country will reflect institutions independent of oil. The results in the second column use the polity measure before known commercialization, where we use a broad definition of known commercialization, dating it as the first year we see oil production in our data base unless we were able to find evidence of earlier commercialization online and in Alexeev and Conrad (2009). The results in column three use a narrower definition, dating commercialization based only on outside sources. The results in column four use the settler mortality measure developed by AJR. 16 The pre-commercialization instruments are valid if the polity measures are persistent over time. In that case, the instruments will be correlated with the time-persistent components of the polity measures while post-commercialization impacts of oil on polity are uncorrelated with the instruments. AJR argue that settler mortality correlates with settlement patterns in European 14 We have also estimated specifications using Kilian (2009)'s real demand variable, as he argues that it is a much better predictor of commodity demands than world GDP. The coefficient on polity was slightly smaller in absolute value, although still precisely estimated to be different from zero. 15 The equation we estimate is a reduced form of the supply and demand equations under certain assumptions about the relationship between income, supply and demand elasticities. 16 Specifically, our instruments are the log of the settler mortality rate from Acemoglu, Johnson and Robinson (2001) and a dummy variable for European countries (for whom we assume that the log of settler mortality is zero). Based on Albouy (2009), we have estimated specifications omitting the nine countries in our data for which mortality estimates were based on soldiers on campaign, and our results are very similar. The 2SLS coefficient on polity is -.0124 (standard error = .0057). colonies and the subsequent political institutions that developed. 17 In all cases, the first-stage coefficients take the expected sign (positive for pre-commercialization polity and negative for settler mortality) and the F-statistics suggests that the instruments have good explanatory power.
All three instruments are only available for a subset of our countries. With the settler mortality instruments, we are mainly missing information from the Middle Eastern countries, which account for ten of the fourteen that are missing. The broad polity variable is missing for Indonesia and Qatar, while the strict polity information is missing primarily for African and Asian countries. OLS results based on the same subset of the data are reported in the panel below the 2SLS results.
When we use the broad pre-commercialization polity instruments, the two-stage least squares result suggests a smaller impact of polity on volatility compared to the equivalent OLS specification. This could reflect a resource curse phenomenon if higher exogenous volatility in oil production facilitates more autocratic governments, perhaps because an autocrat is better able to smooth oil revenues over boom and bust periods. By contrast, when we use the settler mortality instrument, the coefficient estimate more than doubles compared to the OLS specification. This could reflect persistent measurement error in the polity measure. 18 In all three cases, the 2SLS coefficient estimate is negative, and it is statistically significant at the ten percent level or better for the two more precisely estimated instruments.

C. Alternative Political Conditions Measures
A similar story emerges when the polity variable is replaced by other measures of a country's internal political conditions. We consider four additional measures:

Freedom House Political Rights (PR it ):
Freedom House ranks countries on a variety of scales including political rights and civil liberties (next index). We use data from the 2008 Freedom House Survey. 19 The political rights index is based on a checklist of ten questions that is converted to a seven point scale ranging from 1 (most political rights) to 7 (fewest political rights). For consistency with our other measures, we rescale the index to range from 1 (fewest political rights) to 7 (most political rights).
17 Du (2010) provides empirical support for this instrument in an analysis of currency and real economic crises. 18 The analyses of the effects of institutions on macroeconomic conditions generally find that instrumenting for institutions increases their importance, which they interpret as evidence of measurement error (see, e.g., Acemoglu, Johnson and Robinson (2001)). 19 This database is available on-line along with a description of the methodology at http://www.freedomhouse.org/template.cfm?page=15&year=2008.

Freedom House Civil Liberties (CL it ):
Based on the answers to fifteen questions, a score between 1 (most civil liberties) and 7 (fewest civil liberties) is awarded. As with the civil liberties index, we rescale the index to range from 1 (fewer civil liberties) to 7 (most civil liberties). Results are reported in Table 4. Each panel of the table reports univariate OLS and 2SLS

ICRG Expropriation Risks
specifications, equivalent to both the upper and lower panels of the second column of Table 3.
The results for both of the Freedom House variables are very similar to those in Table 3, suggesting a negative relationship between production volatility and internal conditions in both the OLS and 2SLS specifications, and all results are statistically significant at the ten percent level except for the 2SLS specification for the Freedom House Political Rights variable, which is significant at the 11% level. 20 In fact, even though the Freedom House and polity variables are constructed using different methodologies, they are highly correlated (σ FH Civil Liberties -Polity = 0.95 σ FH Civil Liberties -Polity = 0.92).
Both ICRG measures are negatively and statistically significant predictors of oil production volatility, and this holds for both the OLS and 2SLS specifications.
For three of the four measures, the estimated coefficients in the 2SLS regression are smaller in absolute value than the OLS coefficients. The ICRG risk of expropriation is the one exception to this pattern. The first-stage coefficients are positive and significant in all four cases, although the F-statistic suggests weak explanatory power using the ICRG risk of expropriation. Table 5 reports results that include variables that capture cross-country differences in additional, nonpolitical, determinants of development. Our objective is to discern whether the strong relationship between political conditions and oil market volatility is mainly capturing an intermediate relationship between political conditions and another factor, such as financial 20 We also estimated 2SLS specifications using the settler mortality variable as an instrument. The second stage coefficient estimates, equivalent to the four reported in Table 4, were all statistically significantly negative, although the instrument is weak, with first-stage F-statistics below 7 in all specifications.

D. Additional Covariates
openness. We do this by estimating versions of equation (16) that include several additional covariates described more fully below (represented by Add. Cov. i below).
If there is an intermediate relationship between political conditions and the additional covariate, then β 1 from equation (16)  Liberties and the ICRG Composite Index, respectively. Each column in a panel reports results from a separate specification. All specifications include a control for the land area in a country (square miles) on the hypothesis that larger countries may be able to diversify production across sites and so control volatility. This hypothesis bears out as countries with larger land area appear to experience less production volatility. To save space, we do not report the coefficient on the land area variable, but in all specifications, it is negative and statistically significant at the five percent level or higher. 21 The economic impact is small, though, as the coefficient suggests that volatility falls by one-fifth of a standard deviation in countries with four million square miles more land area, which is more than a one standard deviation change in land area among our countries. The first columns in Panels A-D, based on the same time period as the specification in the first row of the left-hand column of Table 2, only include the land area variable, and are included for purposes of comparison to the rest of the table.
In the second column, we report results that include a measure of the country's financial openness: its gross stock of foreign assets and liabilities as a share of GDP. 22 We include this variable to test the hypothesis that political institutions affect firms' ability to produce efficiently at least in part by altering their access to foreign capital. 23 Also, previous work has pointed to the importance of financial openness at the industry level. For example, Rajan and Zingales 21 The reported coefficients in Table 5 are not sensitive to the inclusion of land area. 22 The data are described in Lane and Milesi-Ferretti (2006), and Kose et al. (2006) summarize the pros and cons of different financial openness measures. We have tried other variables, such as the Chinn-Ito Index (see Chinn and Ito (2006), and they are similarly insignificant. 23 The literature on the relationship between financial openness and growth has recognized the potential relationship between political conditions and financial institutions (see, e.g., Eichengreen, Gullapalli and Panizza (2009)). (1998) show that manufacturing industries that are more dependent on external finance grow disproportionately faster in countries that are financially open. While that paper's focus on manufacturing startups may appear irrelevant to the oil extraction industry, startups may provide many valuable support services to established oil extraction firms.
If political institutions affect oil market outcomes by influencing the degree of financial openness, we would expect the coefficient on financial openness to be negative and the coefficient on polity to fall in absolute value. In fact, in all four panels, the coefficient on the financial openness measure is positive, suggesting that countries with more financial openness are more volatile. It is not significant at conventional levels, and the coefficient on the polity variable is attenuated, but not appreciably so. The raw correlation between polity and financial openness for our countries is -.076. This result could reflect the fact that even in countries with considerable capital flows, foreign investment in the oil industry is limited. It may also simply reflect the fact that countries with considerable oil wealth are diversifying by investing abroad (e.g., Saudi Arabia and other major Arab oil and gas producing states), which makes them appear more financially open.
The next set of specifications includes a dummy variable for countries that are in OPEC over the entire time period. Political conditions could drive OPEC membership, as, for instance, countries with institutions that support strong antitrust laws will not be in the cartel. Including the dummy also helps us assess whether the results in the first column are simply identifying differences between the large OPEC countries and the rest of the world or, alternatively, whether polity scores have an effect on production volatility even within OPEC and non-OPEC countries. 24 In principle, it is not clear what the sign of the OPEC dummy would be, as OPEC membership could provide production discipline that reduces volatility. On the other hand, OPEC countries may adjust production to achieve price or profit goals in ways that could contribute to increased production volatility. In general, OPEC membership seems to be associated with high volatility, although the coefficient on the dummy is only significant in one specification. The coefficients on the political conditions variables are generally attenuated, suggesting that our results reflect, in part, the fact that countries with poor political conditions 24 We have also estimated specifications that allow for a quadratic relationship between polity and volatility and found little evidence of important nonlinearities. are in OPEC, but the coefficients on the political conditions variables are still negative, and significantly so in the case of polity, even with the OPEC dummy.
We also included an indicator variable for countries in which there were overt nationalization acts in the oil sector. Specifically, Guriev, Kolotilin and Solin (2009)  The fourth column includes a country's average GDP per capita between 1965 and 2007.
As we have discussed, previous work has found a strong, causal relationship between political conditions and economic development levels, as measured by GDP per capita, among other things. This appears true for the oil-producing countries in our data. For instance, the raw correlation between the political conditions variables and GDP per capita is strongly positive: The results for the polity variable, on the other hand, accord with expectations. The coefficient on GDP is negative, though imprecisely estimated, and the coefficient on polity falls slightly in absolute value. This result suggests that, at least for polity, some of the same factors that hinder economic development contribute to higher oil market volatility. Taken together, the result in Panels A-D indicate that political conditions affect production volatility after controlling for the country's development level (as measured by per capita GDP). This buttresses the conclusion that political conditions can have real effects on production decisions in a particular market.
Finally, we include a variable to measure countries' legal environments. These specifications are not directly analogous to the others in Table 5, as political conditions and legal environments are generally believed to arise independently. Previous work has evaluated "property rights institutions", which protect citizens from governments (and are generally measured by our political conditions variables), and contracting institutions, which facilitate commercial transactions between citizens, as competing explanations for different economic development levels. Measures of contracting institutions include variables such as the number of legal procedures required to settle the collection on an overdue payment. 26 La Porta et al. (1998) show that countries in which the legal system derives from the French (civil-law) tradition have weaker contracting institutions (e.g., higher number of procedures) than countries with legal systems derived from the English (common law) tradition. Acemoglu and Johnson (2005) show that contracting institutions matter little for macroeconomic growth once differences in political institutions are accounted for. We find a similar result for oil production volatility. 27

E. Alternative Specifications
The regressions reported above use average political conditions measures over the sample, so the coefficients are identified by cross-country differences in political conditions.
Our instruments pick up only those components of political conditions which were predetermined before oil was a major factor in a country's economy, but our exclusion restriction will be violated if there is an omitted variable correlated both with poor political conditions and volatile production. For instance, if there is something about dominant religious institutions in a country that leads both to strong, unchecked executives and to volatility in production, our results cannot be interpreted causally. In this section, we focus on variation in the political measure as the source of instability in oil production.
We estimated several specifications similar to those reported in Tables 2 through 5  An alternative specification replaces the polity measure with the change in polity defined as the difference between the average value across the last third of a country's observations and the average of the first third of observations. The production volatility measures are also redefined as the difference between the volatility for the last third of a country's observations and the first third of observations. The results are small and indistinguishable from zero. In sum, there appears to be too little variation in polity scores within countries over time to identify the role of changes in political conditions on oil production volatility.

F. Political Conditions and Other Oil Production Outcomes
Overall, our cross-sectional results suggest a robust causal relationship between political conditions and oil production volatility. To further explore the role of political institutions, Table 6 reports results from a specification where we examined the volatility in the number of wells in a country. The coefficients on polity suggest that the number of wells is also more volatile in countries with poor political conditions, consistent with the hypothesis that poor political conditions impact infrastructure investment and supply decisions. Put differently, this helps confirm that the results in Tables 2 through 5 do not reflect a spurious correlation between physical or other determinants of oil production and political institutions, but instead can be interpreted causally. The 2SLS results are nearly twice the magnitude of the OLS estimates, consistent with the presence of measurement error in the polity measure, and the results are statistically significant at the one percent level. 28 The results in Table 7 speak to whether political conditions affect not just the volatility of oil production but also the absolute level. We consider the mean of production as a share of total reserves by country, under the theory that reserves are exogenous to political conditions and mainly reflect a country's natural oil endowment. The coefficient on polity is positive, suggesting that countries with good political conditions are extracting a greater share of their total reserves over the time period we examine. This could reflect a number of factors. For instance, it could reflect profit-maximizing withholding by monopoly state-owned oil companies 28 Specifications including the OPEC dummy are very similar to those reported in Table 6. in low polity countries, although the result is robust to the inclusion of the OPEC dummy variable. The results are also robust to instrumenting for political conditions. While in no way dispositive, the result is consistent with the hypothesis that poor political conditions impede production.

IV. Implications
Having documented the importance of a country's political conditions to its oil production, we explore how changes in average political conditions of oil producing countries affect price volatility in world oil markets. This allows us to explore whether the results we find at the micro level aggregate up to the macro level. It also allows us to analyze the impacts on volatility of changes in political conditions within countries versus changes across countries in world production shares.
To conduct this analysis we construct several indexes of the political conditions of oilproducing countries over the past several decades. In all cases we construct a political conditions measure according to the following formula (where I it represents one of the political conditions measures described in the previous section for country i in year t and t Π is a global index based on the underlying measure): ω is the share of country i's oil production in world production for year t. We normalize all of the indexes by dividing by the value in the first year that the index is available. The oil conditions index, t Π , is a convenient measure of the average political conditions in oilproducing countries weighted by production.

A. Changes in Production Shares versus Changes in Internal Conditions
What explains the changes in the various indexes? Log differentiation of equation (20) yields: In sum, the data suggest that there has been considerable variation over the past forty years in the political structure of oil producing countries. That change has been driven at times by changes in the political structure within oil producing countries and at other times by changes in production shares across countries.

B. Political Conditions and Macroeconomic Indicators
A natural question to ask is whether the changes over time in the weighted-average political condition have impacted world oil markets. For instance, we might suspect that as production moves away from countries with poor political conditions, which we demonstrated in the previous section had more volatile production, volatility in prices would decline. Table 8 reports results from simple time series regressions which use two of the three indexes depicted in the upper left panel of Figure 3, decomposing the polity measure, to predict oil price volatility. 30 We measure oil price volatility in two ways. In the first and third columns, we measure the within-year variance in the spot price (using data from the Global Financial Database, Commodity Price series). As spot prices are only really meaningful beginning in the early 1980s, we use data from the last 25 years of our sample . In the second and fourth columns, we measure price volatility using the absolute value of the difference between the oil forward contract closing price traded three-months prior to delivery minus the delivery date's closing spot price. (The last daily trade in a month is used as the monthly observation of spot price and as the contract delivery date.) This measure will capture unexpected changes in the market over the short run, which could be driven by supply-side disruptions caused by the political conditions in the oil-producing countries.
The top panel of Table 8  suggest that it takes time before short-run changes in political conditions impact oil production in a country, but that shifts in production to countries with long histories of good governance will lead to lower price volatility.

C. Interpreting Our Results
We have presented a theoretical model that suggests a link between political conditions in a country and the stability of the country's oil production. Empirical evidence suggests that more democratic countries have more stable production over time. Moreover our preferred political conditions index of oil supply suggests a recent downturn in polity among oil producing nations.
One should not, however, interpret our results as suggesting a new rationale for government intervention in energy markets to promote energy security. For example, it would be incorrect to interpret our results as supporting a policy of more domestic supply to reduce oil imports. We say this first because supply shocks of the sort that we investigate in this papercountry-specific and largely idiosyncratic shocks -can in general be quickly replaced in world markets by other suppliers. Thus price shocks are for the most part short-lived. 31 Second, oil is a fungible commodity and price shocks arising from a supply shortfall anywhere in the world affects all oil consumers regardless of the source of their particular oil. As noted by Deutch and Schlesinger (2006) and the National Research Council (2009), among others, a country's vulnerability to oil shocks depends on its consumption of oil relative to the size of its economy rather than its imports.
The results do enhance our understanding of the relationship between political governance and energy supply and more generally of the connections between political institutions and economic performance. In that vein, our paper can be viewed as a contribution towards opening up Acemoglu, Johnson and Robinson's "black box" to see an example where governance affects the workings of a particular market, one that happens to be of particular importance to the world economy.

V. Conclusion
We have analyzed how political conditions in oil producing countries affect the volatility of oil production as well as other oil market outcomes. We show that there is a pronounced negative relationship between the short-run volatility in oil production in a country and its political openness, with very democratic regimes exhibiting less volatility in their oil production than more autocratic regimes. This result holds across several robustness checks including using different measures of political conditions, instrumenting for political conditions and using different measures of oil production volatility.
We next document that the average political conditions of oil producing countries has changed markedly over the past thirty-five to forty years. Using the polity composite democracy index, for example, we find that our oil index falls by 30 percent between 1965 and 1976 and then nearly doubles over the next twenty-eight years. Over the past five years it has fallen by roughly five percent. We then decompose that index into a political index that controls for the distribution of oil production across countries and a production share index that controls for the political structure of producing countries. Again using the polity measure, early changes in our oil security index are driven by the distribution of countries producing oil while latter changes are driven more by changes in political conditions within producing countries.
One should be cautious in drawing policy recommendations from this finding. It does suggest that a research agenda focusing on the role of political institutions in affecting global energy markets is a fruitful one. This paper takes a reduced form approach. Subsequent research will be important to begin to understand the mechanisms that affect supply volatility and what the implications are for policy. Also, our index measures could be used to analyze the potential risks associated with different renewable energy technologies, many of which (for instance, solar electricity and electric vehicles) rely on natural resources that are available in a limited set of countries.

Appendix (Not for Publication). Construction of the Fisher Political Conditions Indexes
We begin by constructing an index of oil security as a weighted average of country level security as measured by one of our political conditions measures weighted by oil production in that year. The index is normalized to equal one in the first year. We describe the process in detail for our polity variable. Fisher (1921) showed that his ideal index satisfied perfect decomposition of an expenditure index into a price and quantity index under conditions satisfied by our data. In our context, a Fisher ideal index provides a perfect decomposition of an aggregate political conditions index into political and production share indexes with no residual: The other political conditions indexes are decomposed in a similar fashion where we simply substitute other country level political conditions measures for the polity measure used here. Robust standard errors in parentheses. All specifications are weighted by a country's average production within the appropriate time period. The first two columns present results where the dependent variable is estimated over the range 1981-2007, where the first year is one year after the data begins since the dependent variables are in first differences. The two right-most columns present results where the dependent variable is estimated over the data range . N = 48, rows 1-3, 5; N=47, row 4 (data on reserves in Cameroon are not available). * -p-value less than 10 percent ** -p-value less than 5 percent *** -p-value less than 1 percent     * -p-value less than 10 percent ** -p-value less than 5 percent *** -p-value less than 1 percent * -p-value less than 10 percent ** -p-value less than 5 percent *** -p-value less than 1 percent Regressions are run using data from 1983 through 2007 (25 observations). Estimated using a Prais-Winsten transformation to adjust for the presence of serial correlation. * -p-value less than 10 percent ** -p-value less than 5 percent *** -p-value less than 1 percent Figure 1