Institutions and the Location of Oil Exploration

We provide evidence that institutions have a strong influence over where oil and gas exploration takes place. We utilise a global data set on the location of exploration wells and national borders. This allows for a regression discontinuity design with the identifying assumption that the position of borders was determined independently of geology. In order to break potential simultaneity between borders, institutions, and activities in the oil sector, we focus on drilling that occurred after the formation of borders and institutions. Our sample covers 88 countries over the 1966–2010 period. At borders, we estimate more than twice as much drilling on the side with better institutional quality. Subsample analyses reveal effects of institutions on exploration drilling in both developing and high income countries, as well as across three types of operating companies. We find that the supermajor international oil companies are particularly sensitive to institutional quality in developing countries. Our findings are consistent with the view that institutions shape both exploration companies’ incentives to invest in drilling and host countries’ supply of drilling opportunities.


Introduction
Institutions, generally defined as 'the rules of the game in a society' (North, 1990), are widely considered to be a fundamental cause of economic growth. Aspects such as constraints on the executive or the rule of law shape the incentives to invest and therefore growth trajectories . However, identifying the causal effect of institutions is challenging because of correlated factors and because institutional characteristics are themselves endogenous equilibrium outcomes (Acemoglu et al. 2005, Besley andPersson 2010).
This paper uses the setting of oil and gas exploration to estimate the effect of institutional quality on investment. 1 Exploration and subsequent extraction requires large up-front capital outlays. A potential investor will take into account the probability of discovery and the net present value of a discovered barrel of oil. The former depends upon geology, the latter upon operational costs and risks, which are again partly determined by institutions. The 'supply side', i.e. how potential host countries facilitate drilling, for example through the licensing process and tax deductions for exploration costs, is also plausibly affected by institutional quality.
For these reasons, oil exploration is expected to vary with institutional quality for a given set of geological conditions. Aggregate figures show that almost 90% of oil investments have historically taken place in upper-middle and high income countries. OECD-countries have discovered about five times more subsoil natural resources per square km than those in Africa. 2 This paper argues that variation in institutional quality between countries can contribute to understanding the uneven distribution of investments and hence the distribution of known sub-soil wealth between countries.
The ideal experiment in our setting would be to treat a given geology with different institutional quality and observe the response in drilling. We mimic such an experiment by implementing a Regression Discontinuity (RD) design. A new global data set including the geocoded location of oil exploration wells and national borders allows us to examine investments 1 We focus in this paper on oil and gas exploration, measured as the number of oil and gas wells drilled. We use 'oil' as a short hand expression for 'oil and gas'. 2 Collier (2010) estimates that OECD countries have discovered 130 thousand USD worth of subsoil natural resources per square km compared to 25 thousand for those in Africa. See McKinsey Global Institute (2013) for the investment figure. 1 in areas proximate to national borders. We consider borders which have not been changed since 1965 and only drilling that took place from 1966. If these borders were determined independently of subsoil oil, a discontinuity in the number of wells at national borders can be interpreted as the causal effect of the border on oil exploration.
We follow the literature on the effects of institutions on economic growth and apply broad measures of institutions: political rights, democracy, autocracy and constraints on the executive branch of government. Previous research has also suggested that these dimensions are important for the likelihood of expropriation of foreign assets (Li, 2009).
Like Michalopoulos and Papaioannou (2013), we place the country with the better score for institutional quality on the 'right hand side' of the border. A border-dummy, taking one on the right hand side and zero on the left hand side, is the explanatory variable of interest in the regressions. The institutional variation is bilateral and strictly ordinal, as we merely use the institutional measure to rank the two countries. Simultaneity between activities in the oil sector and institutional quality would induce a bias in our estimates if it shifted the ranking of the two countries. Institutional quality is therefore measured in 1965 or before, prior to the exploration drilling we consider. 3 To estimate the effect of a given change in institutional quality, we use measures of institutional quality as the treatment variable and the border-dummy as the instrument in a two-stage-least-squares estimation.
Our baseline estimates suggest a jump at the border of 36%, from from 4.1 to 5.6 wells on average. This means that, ceteris paribus, exploration companies choose the better side of the border 58% of the time. Scaled with measures of institutional quality, the estimates imply that a one standard deviation increase in the augmented Freedom House Political Rights Index increases the number of wells by 39%. These are close to long-run estimates, since they are based on the accumulated number of wells drilled in the period 1966-2010.
A rough back-of-the-envelope calculation suggests that the average exploration well in our sample results in production of about 2,500 barrels per day. Based on the 2,100 wells we observe within 10 km of the left hand side of the border, our estimates imply that transplanting the institutions from the right hand side to the left hand side would increase the daily production by an amount that compares to 2% of the global oil production in 2013. The corresponding figure for a distance of 250 km from the borders is 30% of the global oil production in 2013.
Allowing for heterogeneity across companies, drilling by the so called six supermajors of the oil industry (Chevron, Shell, BP, ExxonMobil, ConocoPhillips and Total) is found to be more sensitive to institutional quality than nationally-owned oil companies (such as CNPC, PDVSA, Petrobras or Petronas), as well as the rest of the oil exploration industry. Splitting the sample in terms of income per capita, we find that the effect in general holds, except for democracy and autocracy measures close to borders in the high income countries.
Given that the fundamental causes of economic growth may be classified as 'institutions', 'geography' and 'culture' (see , our contribution is to show that institutions have a positive effect on oil exploration, keeping geography and culture fixed. The effect of institutions itself may be sliced up in three different ways. First, inspired by  who unbundle institutions into contracting institutions and property rights institutions, we show that our effect at least partly works via property rights institutions, although it does not seem to be the only relevant aspect. Second, as a fundamental cause of economic growth, institutions operate through a variety of proximate causes such as investments in physical and human capital. We show that controlling for the initial states of such proximate causes reduces but does not take away the estimated effect of institutions. Third, institutions may affect both demand for drilling by the exploration companies and supply of drilling opportunities by the host countries. Disinvestments from Iran by international investors during the 1990s and early 2000s, due to operating terms and escalating political tension, illustrates the demand side. Mexico banned oil drilling by foreign companies between 1938 and 2013 and serves as an illustration of restricted supply. This may have contributed to the sharp asymmetry in drilling across the US-Mexico border in the Gulf of Mexico. 4 In our setting, we observe only the equi-librium outcome and lack of excludable instruments for demand and supply prevents us from identifying their effects separately.
Our estimates are based on oil exploration close to national borders. There may be a race to drill first at the border, as oil deposits may straddle the border. Although consistent with the hypothesis of this paper, and not posing a threat to the internal validity of our estimates, this would generate an estimate for the effect of institutions that is higher than for oil exploration in general. Empirically we cannot rule out such a competition effect, but show in section 3.6 that this is unlikely to be driving our qualitative conclusion.
To scrutinize the key identifying assumption that national borders are randomly assigned with respect to the underlying geology, we perform a set of tests. We show that the presence of geological basins, where oil can be found, is the same on each side of our borders. Conditional on drilling, there is also no discontinuity at borders in discovery rates. Furthermore, our results are robust to limiting the sample to include "older" borders only (1925, 1938 or 1945), to African countries only, and to so-called 'artificial' states only. The same is true when we exclude offshore borders or borders with interstate conflicts. Finally, the results are robust to the inclusion of previous drilling, previous discoveries or country fixed effects as controls.
The starting point of this paper is the literature on institutions and economic growth , Acemoglu et al. 2001, Acemoglu et al. 2014, Dell 2010, Hall and Jones 1999, Michalopoulos and Papaioannou 2013, Papaioannou and Siourounis 2008and Rodrik and Wacziarg 2005. The paper's contribution lies in the identification of investments as a channel through which institutions such as democracy affect growth. We measure investments as the number of exploration wells drilled. The setting of oil exploration lends itself particulary well to a spatial regression discontinuity design, such as the one used by Michalopoulos and Papaioannou (2013). First, we observe the exact geographic location and timing of each exploration well providing us with rich variation in our outcome of interest. Second, the subsoil nature of oil reservoirs together with the timing of borders make the location of borders plausibly exogenous with respect to oil. The borders we 4 study were typically established before oil had high economic value and before its likely locations were known. In contrast, application of other measures of economic activity used in the literature, such as nighttime lights, is more vulnerable to the possibility that the location of people, firms and political borders are simultaneously determined. 5 Our finding suggests that better scores on measures of democracy, autocracy and constraints on the executive branch of government do increase investments in oil exploration. This is in agreement with Acemoglu et al. (2014), who find a positive effect of democracy on GDP per capita with investments as a channel. They are also in line with , who find positive effects of property rights institutions, such as constraints on the executive, on GDP per capita and investments. In addition, we present evidence indicating that our findings are not only about foreign direct investments. Both domestic and international companies are responsive to the quality of institutions, although the large international oil companies are more sensitive than the other companies.
In the literature on the use of natural resources, Bohn and Deacon (2000) study the effect of expropriation risk on investments in oil exploration and production. Theoretically, they find a negative effect under plausible assumptions, and confirm this empirically in a crosscountry analysis. 6 Our paper improves on their empirical analysis by using micro data on oil exploration, an identification strategy utilising plausibly exogenous variation in institutional quality and more general measures of institutions. 7 Our findings are in line with their findings.
Our finding means that some types of natural capital are economic outcomes. Like human and physical capital, their accumulation (discovery) depends on institutions. This has immediate consequences for the understanding of the determinants of 'Wealth of Nations', as 5 Related issues are discussed by Alesina et al. (2011), Michalopoulos andPapaioannou (2013), Michalopoulos and Papaioannou (2014) and Turner et al. (2014). 6 This finding stands in contrast to the earlier conventional view that weak property rights increase current extraction in expectation of higher expropriation risk and hence lower marginal returns in the future (Long, 1975). 7 The natural resource literature on how resource use varies across levels of development has been concerned with the common pool aspect of natural resources, see for example Bohn and Deacon (2000), Jacoby et al. (2002), Long (1975) and Laurent-Lucchetti and Santugini (2012). A related literature has focused on foreign direct investment under the risk of expropriation, see for example Thomas and Worrall (1994) and Janeba (2002). The literature on oil exploration more generally has either been theoretical or focusing on mature economies like US, UK, Canada and Norway, and has in both strands been less concerned with the institutional setting. See for example Hendricks and Porter (1996), Hurn and Wright (1994), Livernois and Ryan (1989), Mohn and Osmundsen (2008), Pesaran (1990), Quyen (1991) and Venables (2011). measured by the World Bank (2011). It also has implications for the literature on the natural resource curse (see summary by van der Ploeg 2011). Oil reserves and production must be treated as endogenous in empirical models of the effects of oil. This questions identification strategies that assume oil discoveries and production to be driven by geology, without taking into account that they are themselves outcomes of institutions and associated economic developments. 8 The idea that oil exploration, and hence the discovery of sub-soil wealth of countries, depends on the institutional environment may not be accepted by policy makers everywhere.
Countries like Iraq, Nigeria, Venezuela and Angola, who rank among the top twenty largest oil producers in the world in 2012 (EIA, 2013), score relatively low on measures of institutional quality. Promising geology may therefore be thought of as a sufficient condition for oil exploration. In contrast, our results suggest that countries with identical geology but a different institutional setting will fare very differently. To the extent countries can act to improve their institutional environment, they may be in a position to affect the pattern of oil exploration and discovery. We take this to be an important policy message of this paper.
The paper proceeds as follows. The next section describes the data and outlines the identification strategy. Section 3 presents the main results together with sub-sections dealing with the roles of ethnicity and natural geography, the role of property rights institutions versus other aspects of the institutional setting, and the role of a potential common pool effect at the border. Section 4 investigates in detail potential threats to identification. Section 5 examines heterogeneity across different company types as well as between developing countries and high income countries. Section 6 offers concluding remarks. 8 We discuss the implications of endogenous institutions with respect to oil in section 4. See Andersen and Aslaksen (2008), Andersen and Ross (2014), Brunnschweiler and Bulte (2008), Haber andMenaldo (2011), Jensen andWantchekon (2004), Mehlum et al. (2006), van der Ploeg and Poelhekke (2010), Ross (2001) and Tsui (2011) for papers dealing with the resource curse and democracy/institutions. Tsui (2011) and Cotet and Tsui (2013) aim for tackling the endogeneity of oil wealth by using initial resource wealth or discovery rates as instruments and Cassidy (2016) by using information about geological basins. 6 2 Data and empirical strategy 2.1 Data Oil exploration data are provided by the PathFinder database owned by Wood Mackenzie (2011) which covers more than 100,000 individual wells in over 120 countries. This proprietary database approaches comprehensive global coverage and is, to our knowledge, the largest collection of the world's exploration wells. It includes a wide range of country and operator reported information assembled historically and updated on a quarterly basis. We utilise information on the exact location of each well, the year when drilling started, and whether the well turned out to be 'dry' or resulted in a discovery. 9 Figure 1 shows the global distribution of wells available. 10 Data for onshore national borders are from the GADM database of Global Administrative Areas version 2.0 (Hijmans et al., 2010) and for offshore maritime borders the EEZ Maritime Boundaries Geodatabase version 6.1 (Claus et al., 2013). 11 We only include borders which have not changed since 1965, according to . 12 We define our unit of observation as bins of 0.1 km width, stacked from a given national border. We use the distance from the middle of that bin as the distance to the border. We calculate the distance from each well to the closest national border and sum the number of wells drilled in each bin. This sum of wells is our dependent variable. Bins without any wells 9 We define dry wells (versus those recorded as a discovery) using the industry standard definition: a dry well or dry hole in our data is denoted as such where evaluated to contain insufficient oil for commercial production. Wood Mackenzie analysis supplements the raw well data to include post-drilling well evaluation, allowing for a more accurate picture of the eventual result and assessment of the well -and not just the recorded result at the time of drilling. 10 We focus in this paper on oil and gas exploration. For solid minerals and metals such as coal, iron ore, gold and copper, we do not possess exploration information equivalent to exploration wells. To run a similar study for such commodities, we would have to rely on data on actual mines, i.e. projects that have moved into the production phase. Furthermore, many of these commodities have held value far back into human history and geologically they are often located closer to the surface than oil and gas. The presence of such commodities may therefore contaminate the location of borders. 11 All geographic data, including distance and area calculations uses the Cylindrical Equal Area projection. This projection has the advantage of minimal distortion for global projections, while preserving consistent area. We calculate straight-line distances from individual exploration wells to nearest borders and nearest neighbours using the 'near' point-to-line function of ArcGIS.
12 See the online appendix B.2 for additional discussion of our borders data. 7 will not be represented in our main specifications. of institutional quality we turn to the widely used Polity IV database. 14 Democ and Autoc are measured on a 0-10 scale, with higher score indicating more democratic or more autocratic institutions, respectively. A composite Polity score is given by Polity = Democ -Autoc, measured on a -10 to 10 scale. ConEx captures institutionalized constraints on the Executive and is measured on a 1-7 scale, with higher score corresponding to more constraints. 15 Country-level 13 The data are available from 1950 and can be found at: https://www.aeaweb.org/articles.php?doi=10.1257/aer.98.3.808 14 The data start in 1800 and are taken from the 'Polity IV: Regime Authority Characteristics and Transitions Datasets'. They can be downloaded from: http://www.systemicpeace.org/inscr/p4v2012.xls. We follow  and treat 'interregnums' as missing values for all the Polity IV variables. 15 We measure institutions at the country-level and assume that the measures are representative for the institutional quality also close to borders. The issue of dual-institutions pointed to by Michalopoulos and Papaioannou (2013), where the institutions of the 'modern sector' do not penetrate fully to the "countryside' and hence potentially weakens the discontinuity in institutions at the border, is unlikely to be an issue in our setting. Investments in the oil sector are potentially large contributions to the domestic capital formation and likely to attract the attention of the central authorities; laws, policies and procedures relating to the oil sector are likely to be centralised and commonly applied across the country (Daniel et al., 2010). 8 data on country size and landlocked status are from CEPII. 16 We use all countries available, except in section 5.2, where we split the sample into developing versus high income countries. 17 We exclude all formally 'disputed areas' in the offshore data, given the unclear and contested ownership of such areas. 18

Empirical strategy
Our conjecture is that institutional quality affects expected profits in the oil sector and an investment location decision will therefore be determined by institutions as well as geology. Three particular sources of endogeneity may induce bias when estimating the effect of institutional quality on oil exploration. The first is that the geology (the actual likelihood of discovering oil) is not observable to us. The second is that institutional quality is itself an outcome that can be affected by the investment activities under study. The third is that the territory of countries may be correlated with both geology and institutional quality. We seek to overcome these three identification challenges by taking advantage of the location of oil exploration relative to national borders and by utilising the particular sequencing of events in our setting.
Data on the precise geographical location of oil wells and national borders allow for a regression discontinuity design, as potential geological deposits of oil often stretch across national borders. 19 By studying oil exploration very close to the border, we effectively control for geology. To deal with the potential simultaneity between activities in the oil sector and the institutional quality, we measure institutional quality before the drilling we consider took place. Institutional quality is then predetermined. Finally, to make sure that the location of the national borders is not affected by the oil exploration we consider, we include only countries we know have not changed in shape during our period of oil exploration. 16 The data can be found at: http://www.cepii.fr/anglaisgraph/bdd/distances.htm 17 We use World Bank's definition of developing countries and geographical regions, as of July 2012. For more information, see: http://data.worldbank.org/news/new-country-classifications. 18 Disputed areas are defined by the maritime border geodatabase Claus et al. (2013) 19 See Imbens and Lemieux (2008), Imbens and Wooldridge (2009) and Lee and Lemieux (2010) for thorough explanations and discussions of regression discontinuity (RD) designs. given from pre-historic geological processes, taking place in period 1. In period 2, we posit that national borders were located and institutions were formed. In period 3, exploration drilling for oil takes place. We choose to define period 2 as 1965 and earlier because Weidmann et al. We follow Michalopoulos and Papaioannou (2013) and place a given country on the left or right hand side of the border depending on its institutional quality relative to the neighbour.
The better country in a country-pair is placed on the right hand side. Pooling observations on the left and right hand side, the effect of national borders on oil exploration is estimated by the following regression: The unit of observation is a distance-bin b in country i close to the border with country j. A distance-bin is defined as a 100 meter wide strip following along the border. 21 N is a variable measuring the number of wells drilled in the bin. α is a constant and D is a dummy variable taking the value of one on the side of the border with the better institutional quality, and zero on the other side. The assignment variable, X, is the distance to the border, taking negative values on the left hand side and positive values on the right hand side. The f -function is used 20 See section 2 for more information on the data from . We provide sensitivity checks on this choice in section 4.1. 21 The distance from the border is measured to the mid point of the distance-bin. For example, the tenth distancebin is measured at 950 meters from the border and will include all wells drilled from 900 to 1000 meter from the border.
to pick up the underlying distribution of drilling with respect to the distance to the border and we allow it to differ on both sides of the border by interacting with the border dummy D. Z is a vector of controls for country i and u is the error term. τ picks up a jump in the mean at the discontinuity, i.e. the effect of crossing the national border on the number of exploration wells drilled. Note that institutional quality enters equation 1 only in an ordinal sense, i.e. by determining the ranking of the two countries. Simultaneity between the number of wells and the quality of institutions is therefore only an issue if the ranking is affected.
Control variables should not affect the estimate of τ in a correctly specified RD design, but their inclusion may increase efficiency Lemieux 2008, Lee andLemieux 2010).
For this reason we include the total area of country i, which may be correlated with the length of the border, and the landlocked status of country i, which may affect transport costs asymmetrically across borders.
To estimate how responsive oil exploration is to a given change in institutional quality, we scale the jump at the border with the level of institutional quality by applying standard two-stages least squares estimation, with the border dummy as instrument for the level of institutional quality.
The RD-assumption of equal geology on both side of the border is most likely to be valid close to the border and we focus on 10 km distance from the border. However, we also present estimates for 250 km from the border. We present estimates for different specifications of the f -function and use a second order polynomial as our preferred specification. The density of wells is expected to increase towards the border, as the density of land area increases towards the border (this is confirmed in figure B.2). There are two reasons for this. First, moving away from the border means moving towards the geographic center of a given country. Although the width of each distance-bin is constant, the length of each strip will necessarily decrease as one moves farther into a country (think of a square-shaped country and move towards the centre from all four sides at the same time). Since area size is a quadratic function of the distance to the centre of the country, the first derivative is a linear function of the distance to the centre. This simple geometric fact explains why the area size increases as one approaches the border. Second, all countries in the sample are represented close to the border, while only large countries are represented far away from the border. 22 This imposes curvature, i.e. the area size increases in a convex manner as one approaches the border. A final reason for the shape towards the border may be competition for oil close to the border, as discussed in section 3.6.
Gelman and Imbens (2014) warn against high-ordered polynomials and our judgement is that a second order polynomial is appropriate in our setting. 23 As we measure the distance from the border discretely (for each 100 meters), we follow Lee and Card (2008) and cluster standard errors on distance-bins in all regressions. We show that our results are robust to accounting for spatial correlation using the GMM-method of Conley (1999) as implemented by the program of Hsiang (2010).

Results
This section starts with a look at the raw data and a graphical presentation of the jump in exploration wells observed at national borders. We then present econometric estimates of the size of this jump. To get an estimate of the responsiveness of exploration drilling to a given change in institutional quality, we scale our econometric estimates with the level of institutional quality.
We do this for several standard measures of broad institutional quality, which also serves to demonstrate that our finding is robust to alternative measures of institutional quality. Then follows a discussion of the role of geography and culture, the two competing fundamental causes of economic growth, as well as a discussion of unbundling the effect of our broad measures of institutions into different sub-aspects of institutions. The section ends with a comparison of the effects close to versus further away from the border.  Table 1 presents the likelihood of drilling together with the mean and total number of wells on each side of the border. In a given 100 meter-wide distance-bin within 10 km of the border, there is a 8% likelihood that a well is drilled on the left hand side (worse institutional quality) compared to 12% on the right hand side (better institutional quality). The mean number of wells are 2.6 versus 3.0, i.e. 18% more wells on the side with better institutions. In the remainder of the paper, we will focus on the number of wells given that drilling has occurred (the 'intensive margin').

Graphical evidence of discontinuity at the border
The RD design offers visual evidence of the discontinuity in exploration activity at the national borders. Figure 3 shows the number of exploration wells drilled within bins of distances from the border. We follow the standard in the RD-literature and fit separate polynomials on each side of the border. There is a clear jump at the border, with higher likelihood of a well being drilled on the side of the border with relatively better institutions (i.e., the right hand side of the border, defined as the country with a bilaterally higher score on the augmented Freedom House Political Rights Index in 1965 and earlier). 24 Close to the border, the geological likelihood of discovering oil is the same on both sides and, given the assumption of predetermined 24 Note that we use as our baseline measure of institutional quality an augmented version of the Freedom House Political Rights Index measure provided by Acemoglu et al. (2008). This is normalised to a zero to one scale, where one means a country comes closest to the political rights ideals suggested by a checklist of questions. borders and institutional quality, the discontinuity indicates the causal effect of differences either side the border. riod and have recorded associated changes in drilling activity. Ghana, for example, saw strong improvements in their democratic governance scores from 1990s onwards and have seen increased and high sustained drilling activity since the late 1990s, leading to major oil discoveries in 2007 onwards and now a nascent oil sector. Thailand similarly saw progress on democratic institutions in the 1990s and associated increases in drilling activity. Venezuela's shift away from democracy since the 1990s has been accompanied by an associated decline in exploration drilling.

Econometric estimates of the effect of crossing the border
To formally estimate the effect of crossing the border on oil exploration, we follow the approach outlined in section 2.2. Column 1 is based on 5 km distance from the border, whereas column 2-6 are based on a 10 km distance from the border. Column 1 and 2 are without any controls and polynomials in the distance to the border. Column 3-6 include controls for area size and landlocked status at the country level. Column 4-6 include polynomials of increasing orders, estimated separately on each side of the border.  Acemoglu et al. (2008). Only developing countries included as host country, while neighbours may be either developing or high income countries. Onshore and offshore wells. Country-distance-bins without observed wells are exclude.τ/N le f t is the ratio between the estimated coefficient on the border dummy and the mean on the left hand side of the border. This mean is estimated conditional on a polynomial in distance of the same order as in the regression with the border dummy.
The first row in the lower part of the table, indicated withN le f t , provides the mean of the dependent variable on the left hand side of the border. The row below presents the ratio between the estimated coefficient and the left-hand-side mean. 25 The estimates suggest a jump at the border of 0.5 to 1.9 wells, translating into a relative jump of 18% to 40%. The area and landlocked controls increase the estimated coefficients, as do higher orders of the polynomials. Our results are in line with the literature on institutions and economic growth, which has found that differences in institutional quality helps to explain the differences in economic performance between countries. In particular, a positive effect of democracy on economic growth has been found by Acemoglu et al. (2014), Papaioannou and Siourounis (2008) and Rodrik and Wacziarg (2005). Our results suggest that oil exploration may not be different to other economic activity in its relation to institutional quality measured as democracy.

Scaling the effect of the border with institutional quality
The estimate of the effect of crossing the border as presented above had the benefit of relying only on ordinal information in the institutional measure. In this section we take one step further and present estimates based on a cardinal interpretation of measures of institutional quality. Although transforming something as complex as institutional quality into a cardinal score may be a challenging task, a cardinal interpretation is not uncommon in the literature (e.g., Acemoglu and Johnson 2005). We scale the border-effect by estimating with 2SLS, as described in section 2.2. The border dummy D is used as an 'instrument' for the institutional quality I. This serves to take into account how much the institutional quality actually jumps at the border. institutions as well as other aspects. The results presented in this section is an average effect across different types of companies and countries. In section 5.1, we investigate heterogeneity across company types and country groups (developing and high income). 26 As an example, this is found byγ * std.dev./N le f t = 5.941 * 0.27/4.12 27 ConEx has been used by several other authors, e.g. Acemoglu and Johnson (2005) 28 For autocracy, a higher score means more autocracy, so we expect a negative sign. Given that we have defined the left hand side of the border as the side with the lower score, the left hand side is in this particular case the side with the 'better' institutional quality. The relative jump from the 'bad' to the 'good' side based on the autocracy measure is therefore 70%: the mean on the 'bad' side is 6.68-2.76=3.92; and the relative jump therefore 2.76/3.92=0.70.
There has recently been a debate in the resource curse literature on whether oil has a detrimental effect on democracy. 29 Institutional quality and oil exploration (and then wealth and subsequent production) are potentially determined simultaneously. We have argued that spatial variation together with the use of predetermined borders and institutions allow us to identify the causality running from institutions to oil exploration (see also section 4.4). For the literature studying the causality running from oil to political (and economic) outcomes, our finding calls for caution. There will be a positive bias in a regression of institutional quality on some oilmeasure correlated with exploration if the simultaneity is not tackled properly. Cassidy (2016) finds such a bias in his work on the long-run effects of oil wealth on development.

Institutions versus geography and culture
The fundamental causes of economic growth are generally classified as either 'institutions', 'geography' or 'culture' (Acemoglu et al. 2005, North andThomas 1973). In this paper we focus on the effect of institutions and our identification strategy seeks to isolate this from potential effects of geography and culture. We do not attempt to investigate the roles of proximate causes of economic growth, like the accumulation of physical and human capital, the level of technology and the organisation of production. They are considered to be the channels through which the fundamental causes may operate and may be the subject of future work. 30 Geological prospects, equivalent to a form of geography in the North terminology, is unlikely to induce bias in our setting. The idea behind the RD design is that the geology is effectively identical at the border. We present in section 4.1 evidence in support of no disconti-29 For example, Haber and Menaldo (2011), Andersen and Aslaksen (2008), Ross (2013), Cotet and Tsui (2013) and Tsui (2011). Haber and Menaldo (2011) argue that 'increases in resource reliance are not associated with authoritarianism'. In contrast, Andersen and Ross (2014), using the same data, find that oil wealth became a 'hindrance to democratic transitions after the transformative events of the 1970s. ' Tsui (2011) finds that 'larger oil discoveries are causally linked to slower transitions to democracy.' 30 Glaeser et al. (2004) argue that 'human capital is a more basic source of growth than are the institutions.' Note that infrastructure is often purpose-built for the oil sector, and know-how, labour and capital required for oil exploration and extraction are likely to be imported, especially in developing countries. As an example from a developed country, the import-share for equipment to the oil industry in Norway was close to 100% in the first decade of oil production (1970s) and had dropped to just below 40% in 2013, according to the Norwegian Central Bank (governor's annual address, February 14 2013). Rodrik et al. (2004) focus on economic integration alongside geography and institutions. In our analysis, neighbouring country-pairs share bilateral integration. Integration across other borders should to some extent be captured by country fixed effects in column 6 in table 9. 20 nuity in geology at our borders. We also show in section 4.2 that the likelihood of discovering oil, conditional on drilling, is the same on each side of the border. Our finding is furthermore robust to making the geographic areas around the borders very local and less local (we present estimates for 5 km, 10 km and 250 km from the border). Controls for area size and landlocked status take into account that operational costs may be asymmetric across the border due to nonlocal geographic features. Finally, the finding is also robust to the inclusion of country fixed effects, which controls for unobserved country-wide geographic features (see section 4.1). Omitting culture could bias our estimates if: culture changes sharply at national borders; the bilateral ranking of countries in terms of culture is correlated with the ranking based on institutional quality; and culture is correlated with oil exploration. 31 Geo-coded data on ethnicity and the fact that ethnicity often stretches across national borders (Michalopoulos and Papaioannou, 2014), allow us to take into account at least one aspect of culture by including ethnicity fixed effects. In table 4 we split our distance-bins according to ethnicity and the unit of observation becomes then an ethnicity-specific distance-bin located in country i close to the border of country j. We exclude offshore wells, since ethnicity is only relevant onshore, and all wells in areas without specified ethnicity. 32 As a reference point, the first column presents results with our standard specification. Column two includes ethnicity fixed effects, hence the border effect is estimated with variation within a given ethnicity only. Column three and four modify column one and two by the inclusion of second-order polynomials in distance to the nearest ethnicity boundary. This distance is the mean distance between the wells and their nearest ethnicity boundary for each unit of observation. We allow these second-order polynomials to be different on the left and right hand side of the national border. The estimated coefficient on the border dummy is somewhat smaller when ethnicity is controlled for, but the finding of this paper holds in all specifications. In the strictest specification, crossing the national border towards the side with better institutional quality increases the number of wells by 55%, against 95% without ethnicity controls. To the extent ethnicity captures culture, the results in table 4 provide support that our results are not driven by culture.
A final remark is that our design is not well suited for investigating the effects of geography and culture. Instead of falsifying their role, our finding is that institutional quality makes a substantial difference in attracting investments in oil exploration.  unbundle the effects of institutions on GDP per capita and investments, as well as other outcomes, into contracting and property rights institutions. As one of their measures on property rights institutions, they use the Polity IV measure constraints on the executive. Column five in table 3 shows that a better score on this measure increase exploration drilling. This is consistent with the theory of Bohn and Deacon (2000), as investors 32 To measure and locate ethnicity we use the Geo-Referencing of Ethnic Groups (GREG) dataset of global ethno-linguistic regions . This dataset is based upon on maps and data drawn from the Soviet Atlas Narodov Mira -a project charting ethnic groups carried out in the 1960s. The Atlas is notable as the basis for the widely utilised Ethno-Linguistic Fractionization index (Taylor and Hudson, 1972). react negatively to expropriation risk. Measures of property rights and contracting institutions may be correlated, however, and we include in table 5 the border dummy defined on the basis   of property rights institutions together with the four other measures of institutions from table   3. The relative jump due to constraints on the executive is reduced from 71% to 12-16%. The relative jump due to polity, democracy and autocracy from the Polity IV database are reduced by 7-10 percentage points, whereas the relative jump due to the Freedom House democracy index increases with six percentage points.  use UK legal origin as an instrument for contracting institutions and population density in the 1500s as well as settler mortality as instruments for property rights institutions. They find that the instruments enter in a separable form in the first stages, e.g. the legal origin affects contracting institutions only. We do not find first stages with such separable forms in our setting, inhibiting a precise unbundling into contracting versus property rights institutions. The results in table 5 suggest that there is a separate effect of constraints on the executive. To the extent the measure of constraints on the executive soak up all aspects of property rights institutions, the results in table 5 also suggest sizeable additional effects of other aspects of the institutional setting on oil exploration.

Effects away from the border
The effect of institutional quality on oil exploration may be different at national borders compared to elsewhere in the country. Unitization, i.e. 'the joint development of a petroleum resource that straddles territory controlled by different companies', is often regulated by law within countries, e.g. in several states in the U.S. (Boyce and Nostbakken, 2011). This is less common across international borders, although examples exist, for example in the North Sea.
It is less common still between developing countries. The consequence may be a race for oil deposits located close to national borders Wiggins 1984, Libecap andWiggins 1985).
The geology may also be different in border regions than elsewhere. Although the geology may be identical on each side of the border, we cannot exclude the possibility that border regions are associated with geology that differs systematically from the interior of countries.
For example, borders could be commonly drawn in rivers or between mountains, which may in turn be associated with relatively better or worse oil prospects than the country on average.
We assess that these concerns may affect the external validity of our estimates, but not their internal validity. Table 6 investigates how the effect changes as we move away from the border. We first expand the distance from the border to 250 km. Column one includes all distance-bins. In column two, we use a 10 km 'thick border', i.e. 5 km on each side of the border are excluded.
In column three to eight, we increase the thickness to 20 km and then by 40 km for each column, up to 200 km. The idea is that the resource competition effect should be less relevant the further away from the border we get. The effect in column 1 is 52%. From column two to eight, it increases monotonically from 49% to 177%. 33 Although we cannot rule out the presence of a resource competition effect or special geology in border regions, it does not seem to be the whole story: the effect of institutional quality on oil exploration holds qualitatively also away from the border. The exact size of the estimates in this paper should, however, be interpreted with the caveat of possible special circumstances in border regions. On the other hand, the bias in the RD-estimate should be the smallest at the border, where it is most plausible that the underlying geology is the same on both sides of the border.

Validity of the identification strategy
In this section we examine potential threats to the internal validity of our estimates.

Exogenous borders
The formal designation of offshore borders has to a large extent been a latter-half of twentieth century phenomenon, in contrast to onshore borders which have existed in various legal forms for centuries. The location of known and expected oil might therefore be thought to have greater influence on the location of offshore borders. Furthermore, offshore borders are arguably more contestable, since no one lives there and legal claim may be less established. In column one in table 7 we exclude offshore wells and the estimate of the effect of the border is, if anything, 25 larger for this sample of onshore wells only. This suggests that our findings are not driven by strategic location of offshore borders by countries with better institutions than their neighbours. Could it be that our findings are driven by borders moving due to interstate territorial conflicts? Nation states with relatively high institutional quality could have been able to redraw borders via invasion and occupation of neighbouring territories with oil or promising geology.
To challenge our identification, the border would have to be permanently moved, i.e. a lasting occupation and not merely a threat of or attempted occupation. Caselli et al. (2014) argue that the distance from the border to an oil field is a predictor of bilateral wars, but they do not identify many incidents of borders actually moving, nor that any such effect should systematically work in favour of neighbours with better institutions. 34 However, to deal with the potential issue of moving borders, we consider in this paper only stable borders. This means that we always exclude disputed offshore border areas and include only borders which have not changed since 1965 (see explanation in section 2). To deal with instability before 1965, column 2-4 in only and the institutional quality measured before 1966. The finding holds and the size of the estimates are larger in all three cases. The results are also robust to controlling for wells drilled before 1966 (available at request from the authors).
Even for our stable borders, inter-state conflict may affect the likelihood of oil exploration.
Column five in table 7 excludes borders across which there has been at least one incidence of Militarized Interstate Dispute (MID). 35 The estimated coefficient on the border dummy is positive and significant and somewhat larger than when the MID-borders are included. If anything, this indicates that incidence of conflict at a border reduces the appetite to drill close to that border later on.
Could it be that colonial powers were attracted to a given territory because of oil, creating a positive correlation between certain institutional quality and oil drilling? This is not a great threat to our identification strategy as almost all colonization took place long before the location of oil was plausibly known. Furthermore, the colonial powers would need to know very precisely where the oil was located, and draw the border accordingly, since our analysis relies on comparison of areas very close to national borders. Moreover, instead of claiming that colonial powers carefully assigned borders according to, for example, the presence of natural resources, Alesina et al. (2011) and Michalopoulos and Papaioannou (2016) argues that many states in Africa were imposed "artificial borders" by colonial powers. In table 8 we restrict our sample to Sub-Saharan African countries only (column 1), African countries (column 2), countries belonging to top third partitioning countries (column 3) and countries belonging to the third of countries with the straightest borders (column 4). the two latter criteria is based on data from Alesina et al. (2011). We impose that both countries in a pair satisfy the criteria. We present estimates for onshore wells only, as our judgment is that the artificial borders arguments is mostly valid for onshore borders. Results are the same if we include also offshore borders.
As these restrictions reduce our sample considerably, we present estimates for 250 km distance 35 To define borders with MIDs we employ the Dyadic MID dataset used by and Martin et al. (2008). The dataset codes MIDs with a hostility level ranging from 1 to 5, where 1=No militarized action, 2=Threat to use force, 3=Display of force, 4=Use of force and 5=War. We follow Caselli et al. (2014)   What about other unobserved characteristics correlated with the location of national borders, the location of subsoil oil deposits, and the quality of institutions? Mountain ranges, rivers or deserts might be candidates. However, these features do not appear to systematically imply the presence of oil on one side of the border versus another. Furthermore, we are not aware of any evidence that topographic features are correlated with the quality of institutions. We have therefore so far been unable to locate a plausible candidate for such co-determination. 36  The distribution of random points is symmetric with no discontinuity at the border, equivalent to the surface area of basins not jumping at the border, in stark contrast to the distribution of actual oil wells. This provides strong support for the key identifying assumption of this paper: the location of borders has been determined independently of the location of oil. Consistent with a smooth likelihood of finding a geological basin either side of borders, as presented at the end of the former subsection, column one in table 9 presents an insignificant coefficient on our standard border dummy when the discovery rate is used as the dependent variable. 37 The discovery rate is defined as the share of non-dry wells in a given distancebin, i.e. the ex-post likelihood of discovering oil. In contrast to the number of wells drilled, the likelihood of finding oil does not appear to differ across the borders. Although the oil exploration companies have geological information available before they start drilling, there is uncertainty about whether a given well will lead to discovery before it is drilled. It is likely, however, that experience in an area increases the likelihood of discovery. On the other hand, it may be that the most promising wells based on the geological information at hand are drilled first, which would contribute to decreasing the likelihood of discoveries in a given area over time. In column two, we include the number of wells drilled before 1966 as a control variable. 38 The results suggest that previous drilling neither affects the discovery rate nor the effect of the border dummy. In column three we modify our baseline specification by including the number of wells drilled before 1966 as a control variable. In this case, previous drilling serves a similar function as a lagged dependent variable, potentially soaking up unobservable characteristics not otherwise accounted for in our analysis. The effect of previous drilling is positive and significant, which could reflect that additional wells have higher expected value due to, for example, infrastructure already in place. The estimate in column three demonstrates that our results are not driven by some accidental early exploration and path-dependency since then.

The likelihood of discovering oil and path dependency
As common infrastructure could lead to spatial correlation and potentially deflate our standard errors, we present in appendix table B.3 our baseline model with standard errors adjusted for spatial correlation and the result holds. In column four, we include the number of discoveries made before 1966 as a control, resulting in a relative jump of 31%.
As we include only distance-bins with observed wells in our baseline specification, a question may be raised about the continuity of our running variable, the distance to the border. In column five we include all distance-bins, i.e. within 10 km of the border we assign zero wells to distance-bins without observed wells. This increases our sample size tenfold. the relative jump at the border is now estimated to 75%. We therefore conclude that any lack of continuity in the running variable does not seem to bias our estimates away from zero.
Could it be that countries with better institutions are also better at reporting drilled oil wells and our findings are just an artifact of such reporting bias? The PathFinder database from Wood Mackenzie that we use in this paper, is considered to have as comprehensive country 38 For this control variable, zeros are assigned to distance-bins without drilling before 1966.
coverage as any single dataset on oil exploration. It relies in large part on company reporting, but also desk research, country reporting and cross-checking for verification. If underreporting by companies is an issue, our expectation is that this is least severe at international companies. Table 11 shows that our results are robust across different company types, including the six so-called supermajors. In addition, the results also hold if we run the regressions for the subsamples of each company type only (results available at request from the authors). Thus, also within each company type, there is more drilling on the better side of the borders. Note also that the underreporting would have to be quite severe to plausibly alter our results. In a sample of 51 876 wells drilled in the period 1966-2010, we observe 16 979 wells (33%) on the left hand side and 34 897 wells (67%) on the right hand side, based on the Freedom House measure.
Reporting bias at the country-level is effectively controlled for by the inclusion of country fixed effects in column 6 of table 9, and the result holds. Further, as reporting bias is presumably smaller in high income countries, the fact that our findings hold also for this sub-sample (except for democracy and autocracy measures very near borders), is reassuring.
Although we cannot rule out bias from mis-reporting or under-reporting, the results above give us confidence that this is not the entire story.

Controlling for other factors changing at the border
As discussed in section 3.4, we focus in this paper on distinguishing between the effects of institutions on the one hand and geography and culture on the other hand. In this section we seek to illuminate the role of proximate causes of economic growth in our setting. We augment our baseline regression with a second border dummy based on bilateral ranks in terms of GDP per capita, GDP, employment (L), human capital (H), physical capital (K) and total factor productivity (A). Together these factors fully describe GDP, according to a standard production function: GDP = F(L, H, K, A). Table 10 presents the results. Notably, our standard border dummy based on institutions is always significant. As the sample size varies, we reestimate our standard model on the same samples for comparison (see table B.5). Controlling for the border dummies defined based on GDP per capita, H or A reduces the estimates. This is as expected given, for example, Hall and Jones (1999), who found that variation in institutional quality is an important explanation for varying TFP-levels, and hence income per capita, across countries. However, note that the differences in the estimates are very small and we are cautious to read too much from them. The estimated relative effect is in the range 29%-54%, with the main reason for the variation across columns being the sample size, as seen in table B.5. 39 We conclude that the coefficient on the institutions border dummy is remarkably stable. A note of caution to these estimates is that the two border dummies may be determined simultaneously, imposing a bias in our estimates (they are mutually 'endogenous controls'). Notes: Dependent variable is the number of exploration wells for each 100 meter wide distance-bin in country i at the border with country j. The second border dummy is defined according the bilateral rank of the variables indicated in the column headings, based on data from the Penn World Tables (PWT)

Reverse causality
A separate issue to the position of the border is potential feedback from oil exploration and subsequent production to institutional quality. The institutional quality is an equilibrium outcome and natural resources are likely to affect this outcome. One potential channel of the resource curse is the deterioration of political institutions in the face of natural resource rents (Ross 2001, Collier andHoeffler 2009). 40 As the correlation between oil extraction and institutional quality is suggested by most papers in the resource curse literature to be negative, this would likely bias our estimates downwards towards zero and hence work against the finding of this paper. However, to accommodate any feedback-bias, we have throughout the paper followed  and used prior institutions, i.e. institutions are measured before the oil exploration we consider took place. 41 Furthermore, effects of previous exploration should be accounted for by the previous exploration or previous discoveries included in column three and four in table 9.
5 Heterogeneity: company types and countries

Heterogeneity across company types
Oil exploration is undertaken by a range of different types of companies, varying in scale, degree of vertical integration and model of ownership. If these companies face different objectives and constraints, a country's institutional setting may affect them differently: large publiclytraded companies may face high potential reputational costs; state owned companies may be insulated from political risk by their state backers or weigh non-commercial factors more heavily than their private counterparts; small specialised exploration companies may be backed by risk-willing owners and capable of handling high risks. 40 Collier and Hoeffler note that these effects are large: "after 28 years a country with mean income but with resource rents worth 30% of GDP would have a 'checks' score in the 22th percentile instead of in the 34th percentile, and a democracy score in the 25th percentile instead of in the 40th percentile." 41 Recall also that we to define the border dummy use institutional quality simply to rank the two bordering countries. Any feedback from oil exploration and extraction would therefore need to alter the ranking, and not just the level, to induce a bias.

33
The so-called six supermajors (Chevron, Shell, BP, ExxonMobil, ConocoPhillips and Total; indicated IOC6) are international oil companies with predominantly non-state ownership.
They are vertically integrated by engaging in the whole industry value chain from exploration to production to downstream activities. Companies belonging to this group, including their subsidiaries and predecessors, are listed as operators for 16% of the total 112 thousand wells we observe across developing and developed countries. National oil companies (NOCs) are typically set up to secure oil rents accruing to governments and to carry out national strategic objectives. 42 These companies are listed as operators for about 30% of the full sample of wells, and some of them operate outside of their home countries. A third type is the non-supermajor private companies, which include smaller specialised exploration companies who may seek to win particular licenses, engage in exploration and then offload operations to more integrated companies. Anecdotal evidence suggests that the global oil industry previously moved towards less vertical integration and more out-sourcing of high-risk exploration, although that trend may now be reversing post-financial crisis.
In table 11, we augment with interaction terms our baseline model in column 5 in table 2 to test for heterogeneity across company types. Column one includes an interaction between the border dummy and a dummy, COMP, taking one for wells operated by one of the 'IOC6' supermajors. 43 In column two, also the subsidiaries and predecessors of the six supermajors are included in the interaction dummy (indicated IOC6+). Column three does the same for NOCs, while only wells operated by NOCs in the NOC's home country are included in the dummy in column four. Finally, in column five the interaction dummy takes one for wells operated by companies not defined as either IOC6+ or NOC. The coefficient on the border dummy is always positive and significant. The coefficient on the interaction term is relatively large and 42 Many formerly state-owned oil companies have been privatised, particularly during the 1990s. For the analysis in this section we including in the coding of 'OTH' those no longer under majority state-ownership as NOCs in 2013. In all cases we look at the operating company only. Oil fields may be developed with multiple owners and minority participants, and ownership can change over time, especially moving from the exploration phase to the production phase. We thus limit our analysis to the well operator at the point of exploration only. This excludes any examination of the role of non-operator investors or those who may acquire a stake subsequent to exploration (which can sometimes apply to state-participation, where NOCs may acquire minority equity as part of the production phase). 43 Note that the dummy indicating company type is also included separately. positive for the IOC6s and negative for the NOCs and the remaining companies, 'OTH'. These results indicate that all company types are sensitive to institutional quality, but the super-majors more so than the others. Better institutional environments may spur foreign investments, while countries with weaker institutions may rely on home grown NOCs, strategic (foreign) NOC partners or smaller exploration companies.

Heterogeneity across countries
Throughout the paper, we use the sample of all countries for which we have data. We now investigate whether the effect is different in developing countries than in high income countries.
Many of the investments in the oil sector may historically have originated in high income countries. Thus, the information asymmetry issues between source and host countries, known from other work to hamper investment flows, are likely to be more of an issue for developing countries. Developing countries are also known to have less friendly business climates and more cumbersome bureaucracies, also reducing investments. Presumably, the asymmetry and bureaucracy issues are less severe in countries with better institutions. 44   Harding and Javorcik (2011) for a discussion of information asymmetry and red tape issue in the context of foreign direct investments across developing vs. high income countries. In a sample of 51 876 wells drilled in the 1966-2010 period, we find that 24% are drilled by national oil companies (NOCs), 21% by one of the six major international oil company, including their subsidiaries and predecessors (IOC+) and 54% by a rest category. (Smaller) international oil companies are likely to make a substantial share of the rest category. 45 Estimates available at request from the authors. when we use democracy or autocracy measures of institutions. We leave for future research to investigate why the investment behaviour seems to be different at those borders.

Conclusions
A natural experiment of borders, randomly assigned with respect to geology, together with pre-determined institutions allow identification of the responsiveness of oil exploration to differences in institutions. Crossing a national border is found to generate a statistically and economically significant jump in exploration drilling for oil and gas. The number of wells increases by 36% and, all else equal, exploration companies prefer to drill on the side with the better institutions 58% of the time.
This paper contributes to the debate on the drivers of cross-country differences in economic performance. First, it shows that institutions have a strong effect on investments. Second, it reveals that the observed distribution of oil wealth and extraction across countries is endogenous; the natural capital component of the 'wealth of nations' responds to institutional quality.
Some regions of the world, such as sub-Saharan Africa, are likely to be 'under-explored' with respect to geology alone, which can explain why they have found relatively little oil per square km. Endogenous oil exploration also poses a challenge to the empirical analyses of the effects of oil wealth and extraction. Third, for governments it is an important message that promising geology may not be sufficient to attract oil exploration. To the extent they can improve the institutional environment, they may accelerate discovery and increase their country's level of natural capital.

37
A Summary statistics To get a better sense of what the average jump at the border means in terms of oil production, we use production data reported at the field level and conduct a crude back-of-the-envelope calculation. 46 By associating wells with the producing fields that result from successful discoveries, we are able to calculate a measure of the expected daily production per exploration well.
In our baseline sample of wells within 10 km distance to the border, the calculated production per original exploration well is about 2 500 barrels per day. This is based on 503 wells, after the bottom 5% and top 5% are removed to avoid influence from outliers. The same exercise in the 250 km sample suggest an average production of 2 300 barrels per day (based on 4 336 wells). 47 We observe in our 10 km sample of 2396 border-distance bins, 2112 wells on the left hand side and 4724 wells on the right hand side. Our estimate of 36% more drilling on the right hand side due to relatively better institutions (column five in table 2), suggests that the number of wells on the left hand side would increase with about 750 wells. Given an average production per exploration well in the 10 km sample of 2500 barrels per day, this would correspond to about 1.9 million extra barrels of oil production per day, about the production of Norway, or about 35 billion USD per year given an oil price of 50 USD per barrel. 48 In the 250 km sample, we observe 16752 wells on the left hand side and 25435 on the right hand side of the border. We estimate in column one in table 6, moving from the left to the right, a jump of 44%. The corresponding increase in the number of wells on the left hand side is then 46 Our field level data is drawn from the Wood Mackenzie PathFinder database (Wood Mackenzie, 2011). It contains information on producing fields, which can be associated with our various exploration wells. The dataset contains information on annual production volumes from 1965 to 2011 for both oil and gas production, in barrels of oil equivalent. 47 Looking at the medians instead, the figures per well are 500 and 800 barrels per day for 10 km and 250 km in the developing country sample, and 800 and 1 250 barrels per day in the maximum sample of all countries. We conclude that 2 300 -2 500 barrels per day may be reasonable, or perhaps somewhat large. 48 The calculations are: 2112 wells on the left hand side * 0.36 = 760 extra wells. In USD, 760 wells * 2 500 barrels production per day per well * 365 days per year * 50 USD per barrel = 35 billion USD per year. about 11 200, or 26 million barrels per day given a daily production of 2 300 barrels per day.

This corresponds to 470 billion USD per year. 49
To put these figures in perspective, the world's total oil production was 87 million barrels per day in 2013, according to BP Statistical Review of World Energy June 2014. The U.S. production was 10 million barrels per day, while the Middle East and Africa produced 28 and 9 million barrels per day, respectively. In other words, our estimates translate into a jump corresponding to about 2 % of the world's oil production in the 10 km sample and about 30% of the world's oil production in the 250 km sample.
Note that the production level per well depends on investments and is therefore endogenous.
The above calculations should therefore be interpreted with this important caveat in mind.
Note also that the wells studied in this paper are exploration wells and not production wells.
The field data on production necessarily correspond to the production phase rather than the preceding exploration phase. Additional (production) wells are needed to extract oil and gas, and the figures we report are therefore not informative about the total drilling necessary for a given level of production. Furthermore, we divide the production on all exploration wells, not only the non-dry ones. As shown in table A.1, about 56% of the wells drilled in our 10 km baseline sample resulted in discoveries (indicated with the dummy variable 'D = 1 if nondry'). Therefore, our calculated figures measure the production resulting from total exploration drilling, which also best resembles our interest: the jump in production we can expect when moving from the left to the right hand side of the border. 49 The calculations are: 25 435 wells on the left hand side * 0.44 ≈ 11 200 extra wells. In USD, 11 200 wells * 2 300 barrels production per day per well * 365 days per year * 50 USD per barrel = 35 billion USD per year.

B.2 Data on stable borders and maritime borders
We provide sensitivity checks across a range of different sub-samples characterised by the most recent date of border changes. To define stable borders, we utilise a new dataset with timevarying information on country shapes. The CShapes borders dataset from Weidmann et al.
(2010) draws on the Correlates of War project (COW, Barbieri et al. 2008), which captures the territorial changes for the countries of the world. The dataset was refined by Gleditsch and Ward (1999) to include an extended list of independent states. The dataset uses several definitions of territorial change. First, new states can become independent, as in the case of the secession of Eritrea from Ethiopia in 1993. Second, states can merge, as in the case of the dissolution of the (Eastern) German Democratic Republic in 1990 and its ascension to the (Western) German Federal Republic. Third, we may have territorial changes to a state that do not involve the emergence or disappearance of states. CShapes includes all changes of the first two types. For the third type, the CShapes dataset only code changes that (i) affect the core territory of a state, and (ii) affect an area of at least 10,000 square kilometers. For our main specification we use the CShapes dataset taking only countries with stable borders prior to 1965. For additional checks we also use the CShapes dataset for post-1945 borders.
In addition to the CShapes stable borders we extend these cut-off dates backwards using a complementary dataset that extends further back in time. Here we utilise the TerrChange database V4.1 (Tir et al., 1998). The TerrChange database records instances of territory changes, marking them by the country who gained territory or lost territory. This does not contain GIS information on the location of territory loss, but instead records whether the country gained or lost territory, and whether this was to its homeland or dependent territories such as colonial holdings. We construct two (arbitrary) year cut-offs -countries whose border has been unchanged since the years 1925 and 1938; this allows us to generate a reasonably sized subsample while going back further in time reduces our sample size considerably.
We construct the sub-samples by calculating the most recent border change per country as either a gain or loss to that countries homeland territory (we exclude changes to its dependent territories). We then include only those countries who have data on them in the database, and had a border change prior to the year stated (this is a conservative approach as we do not include missing countries, even though they might be missing from the dataset due to no border changes since the start of the dataset in 1816). The lists of countries we end up with is presented in table B.1. The estimation results for the two sub-samples limited to these countries are presented in table 7. Note that other data limit the actual samples in our setting to 8 and 7 home countries for 1938 and 1925, respectively. extending up to the limit of the continental shelf. This approach was subsequently adopted by most countries, and replaced freedom of the seas rights originating in the seventeenth century, such as those extending three nautical miles from a nations coastline -the cannon shot rule. The UN Law of the Sea (United Nations 1958) codified in international law the position of adjacent maritime borders (the median line boundary) and the continental shelf limit (i.e. where international waters begin). A subsequent conference, UNCLOS III 1973-1982fined the definition of how far sovereign rights extend vis-a-vis international waters to give us present-day Exclusive Economic Zones (EEZ) (United Nations, 1982). This later conference also updated the principle of defining adjacent (or opposite) national boundaries, but in practice this would rarely imply deviation from the 1958 median line approach. Resulting EEZs typically extend 200 nautical miles from a nations coastline. For our identification we do not exploit the precise limit of sovereign waters versus international waters but instead the position of maritime borders vis-a-vis a countrys neighbour. While inherently more contestable than onshore borders, the position of these borders typically pre-dates oil exploration due to median line extrapolation agreed in 1958 and earlier bilateral agreements. Furthermore, the exploration for and exploitation of oil and gas beyond shallow coastal waters only began in 1979 with the first deep-water drilling technology deployed to the Gulf of Mexico.

B.3 Extra graphs and tables
Below we present extra graphs and tables referred to in the text. Figure Table   B.5-B.6 relate to section 4.3 and are described in more detail below. Table B.7 presents the countries included in our samples. Figure    Notes: Column one as column five in table 2 and column two as column one in table 6, except we allow for spatial correlation between observations by employing the procedure programmed by Hsiang (2010), which builds on Conley (1999). Standard errors are robust to heteroskedasticity. To measure the distance between observations, we use the average longitude and latitude of the wells drilled in a given country-border specific distance-bin. The distance cutoff is set to 10 km and 250 km, the same as the distance to the border. We use Hsiang's default kernal to weight spatial correlations. For more details about the procedure, see http://www.fightentropy.com/2010/06/standard-error-adjustment-ols-for.html.  Table B.5 is described in the text in section 4.3. Table B.6 includes the control-dummies in different combinations. The logic follows the accounting identity in terms of the production function: Y = F(L, H, K, A). Including Y as a control takes into account L, H, K, and A, which are columns 1 and 2 in table 10. We therefore include only combinations of maximum three out of the four variables L, H, K and A in table B.6. The first column includes the three factors L, H and K. Then the next three columns include different combinations of those three factors. In the three next columns we allow A to substitute each of the factors, L, H and K. In other words, in the four first columns, institutions are allowed to work via A. In the three next columns, institutions are allowed to work via L, H and K, respectively. Across the different specifications, the coefficient on the institutions-based border dummy is stable and statistically significant at the ten percent level in all but one of the seven cases. The estimated relative jump is 27%-29%, hardly reduced from the 30% jump estimated in the model without the proximate causes (column 6 in table B.5).   Figure B.3: National borders and geological basins Notes: Map shows onshore and offshore national borders together with geological basins, structures in which oil can be found, marked as dark areas.