The Shifting Natural Wealth of Nations: The Role of Market Orientation

This paper explores the effect of market orientation on (known) natural resource wealth using a novel dataset of world-wide major hydrocarbon and mineral discoveries. Consistent with the predictions of a two-region model, the empirical estimates based on a large panel of countries show that increased market orientation causes a significant increase in discoveries of natural resources. In a thought experiment whereby economies in Latin America and sub-Saharan Africa remained closed, they would have only achieved one quarter of the actual increase in discoveries they have experienced since the early 1990s. The results call into question the commonly held view that natural resource endowments are exogenous.


I. INTRODUCTION
The literature on economic development often assumes that natural resource endowments are exogenous. The consensus in that literature is that resource endowments alongside institutions or legal origin, and geography play a crucial role in determining economic and social outcomes. 1 In contrast, the resource economics literature has emphasized that the resource base is endogenous to investment in exploration and extraction. 2 That literature has, however, overlooked the role that market orientation and institutions play in driving investments in the resource sector. Our aim is to bridge the gap between these two literatures and explore the effect of market orientation on the discovery of proven (known) natural resource wealth.
Countries with weak rule of law, high political or default risk, underdeveloped financial markets, or high transaction cost and deficiencies in governance may attract only limited investment flows even if they offer high rates of return (Shleifer and Wolfenzon, 2002). 3 Specifically for the natural resource sector, empirical evidence suggest that a stable political environment, a low risk of expropriation, and a favourable investment climate boost investment (Bohn and Deacon, 2000;Stroebel and van Benthem, 2014). 4 5 We present systematic evidence that policies geared toward economic liberalization and increased market orientation lead to major natural resource discoveries that eventually boost extractive activities in those countries. We thus demonstrate that increased market orientation in developing countries is an important determinant of proven natural resource wealth.
The experience of the United States during the nineteenth and early twentieth century provides a historical account of the role of market orientation in driving natural wealth. Although the United States at the time of independence was considered to be a country of "abundance of 1 See Acemoglu, Johnson and Robinson (2002), Easterly and Levine (2002), Glaeser and Shleifer (2002), Hall and Jones (1999), and Rodrik, Subramanian and Trebbi (2002). 2 See Pindyck (1978), Arrow and Chang (1982), and Devarajan and Fisher (1982). 3 The literature linking institutions and international capital flows relates to the so-called "Lucas' paradox" (Lucas, 1990). Policies and institutional factors have been shown to play an important role in explaining the magnitude and nature of capital flows to developing and emerging economies (Alfaro, Kalemli-Ozcan and Volosovych, 2008). 4 Irreversible investments in the resource sector involve sunk costs and are subject to holdup and the political risk of expropriation (Long, 1975). 5 Exploiting variations in market orientation and oil deposits sitting on either side of the border, empirical evidence suggests that institutions substantially affect oil and gas exploration (Cust and Harding, 2017). land but virtually no mining potential" (O'Toole, 1977), by 1913 it was the world's dominant producer of virtually every major industrial mineral (David and Wright, 1997). Rather than being driven by a comparative advantage in geological endowments, this resource-based development of the United States was driven among other things by an open market orientation and an accommodating legal environment with the government claiming no ultimate title to mineral rents (e.g., Wright and Czelusta, 2004). 6 In stock terms, proven reserves of natural resources are today significantly higher in advanced countries than in developing countries (World Bank, 2006). In flow terms, however, we observe a shift in resource discoveries from developed to developing countries over the past decades that coincides with increased market orientation in developing countries. The trend towards more market orientation seems to have led to a shift in the geographic distribution of discoveries. As a consequence, the share of worldwide resource discoveries in Latin America and the Caribbean and Sub-Saharan Africa has doubled over the past decades (see Figure 1).

Figure 1: Number of Natural Resource Discoveries by Region and Decade
Note: Based on data from MinEX for natural resource discoveries and Mike Horn for hydrocarbon discoveries. High income countries include OECD members as well as Bahrain, Brunei, Cyprus, Equatorial Guinea, Kuwait, Oman, Qatar, Saudi Arabia, Trinidad and Tobago, and the United Arab Emirates. 6 Economic historians argue that natural resources may be under-produced due to a lack of effective property rights (e.g. Anderson and Libecap, 2011) and that private mineral rights became more explicit as mine values increased (Demsetz, 1967;Libecap, 1976). With increased competition for valuable resources, informal rules were insufficient to reduce risk and support long-term investment to develop the mines. Making property rights more formal boosted mining investment.
Anecdotal evidence suggests that increased market orientation was followed by increased discoveries across continents and types of natural resources (see Table 1). The increase in discoveries after countries open up to the global economy appears to be quite stark. In Peru, for example, discoveries more than quadrupled, in Chile they tripled, and in Mexico they doubled. In Ghana, discoveries only started to occur after the opening of the economy. To motivate our empirical analysis, we put forward a simple two-region model of endogenous reserves based on Pindyck (1978) where multinational corporations are faced with an implicit tax which proxies for how closed market orientation is, and seek the lowest cost location. The model explores the interplay between market orientation and other channels such as the increase in the marginal cost of discoveries and (demand driven) natural resource price shocks.
In turn, key model predictions of our model are then taken to the data.
For our empirical analysis we build a unique and hitherto unexploited dataset of the universe of world-wide major natural resource discoveries since 1950, covering 128 countries, 33 types of natural resources and over 60 years. Our main explanatory variable is a generic measure of market orientation. We provide OLS results as a benchmark but to account for the endogeneity of the market orientation variable, we then use an instrumental variable approach based on Buera et al (2011): a country's choice to liberalize its economy depends on the policies of neighbouring countries in general, but also on how successful other countries with liberalized and closed economies, respectively, performed. We include country fixed effects as well as year-by-resource fixed effects in our panel estimates to control for time-varying resourcespecific factors such as technological progress in extractions of the different types of minerals Our paper is related to the theoretical and empirical literature on exhaustible resource exploitation and exploration. Natural resource exploration and discoveries have been investigated either as a deterministic or a stochastic process (e.g. Pindyck, 1978;Arrow and Chang, 1982;Devarajan and Fisher, 1982). The canonical model is the exploration model developed by Pindyck (1978) where a social planner maximizes the present value of the social net benefits from the consumption of oil and the reserve base can be replenished through exploration and discovery of new fields. 7 We apply this model to a two-region world to explore the relationship between exploration investment and discoveries where multinational corporations are faced with explicit taxes and implicit taxes (as proxy for lack of market orientation) on their investment in the South but none in the North and arbitrage between different locations.
This paper is also related to the literature on the so-called "resource curse". 8 In particular, empirical evidence suggests that the curse in terms of the effect of natural resources on growth is less severe and can even turn into a boon if the quality of institutions is beyond a certain threshold (e.g., Mehlum, Moene and Torvik, 2006;Boschini, Pettersson and Roine, 2007).
While this literature has long focused on the direction of causality running from natural resource endowments and institutions to growth and conflicts, our empirical results suggest that causality running from policies and market orientation to natural resource endowments is equally important. Our focus is on "upstream" rather than "downstream". As such we are concerned with "external" policies and institutions that are geared toward foreign investors who conduct exploration activities. Our results do not contradict the fact that subsequent to a discovery, countries with poor "internal" institutions (e.g. weak state capacity) may experience poor economic performance and/or social outcomes or even civil strife.
The remainder of the paper is organized as follows. Section II puts forward some predictions based on a simple two-region model of depletion and discoveries, and discusses our empirical strategy. Section III presents the data used in the empirical analysis. Section IV presents the main results and key robustness checks. Section V concludes.

II. ANALYTICAL PREDICTIONS AND EMPIRICAL STRATEGY
We specify a simple two-region, three-period model of endogenous reserves and natural resource discoveries with international resource companies (IRCs) arbitraging to decide on the optimal geographical allocation of their exploration and depletion activities. -Discoveries of natural resources in both the North and the South increase with demand for these resources or with the world price of natural resources.
-A lower stock of in situ natural reserves corresponding to more cumulative discoveries in the past depresses discoveries today, albeit that in the short run we expect the effect on discoveries to be positive due to probing and increases probability of discovery.
-Better geological conditions and a more easily accessible stock of natural resources depress extraction costs, and thus lead to more discoveries and higher depletion rates.
-Higher taxes on IRCs or a less outward orientation for IRCs holds back discoveries of natural reserves.
Other (not mutually exclusive) forces include the rise in global resource demand emanating from emerging economies (e.g., China and India) that prompts more exploration efforts and discoveries and increases in the marginal cost of exploration. In our empirical analysis, we allow discoveries to depend not only on ease of access for IRCs but also on global resource demand shocks and changes in marginal costs of discoveries due to depletion forces. Taxes and restrictive access in the South lead to a shifting of exploration activities and discoveries from the South to the North due to tax shifting causing a rise in world resource prices As the South liberalizes and gets more open to IRCs, the world price of resources falls and exploration activities and discoveries shift from North to South. However, we do not test this last prediction directly but instead allow for the effect of prices and demand shocks on discoveries.
We develop an empirical strategy that allows us to test the above analytical predictions. Our main concern is the estimation of the impact of market orientation on natural resource discoveries. Let , , be the number of discoveries for country at time of resource . Our baseline empirical model estimates a 3-way panel defined by where , is an indicator for whether country is open at time 1, , , is a set of controls which depending on the case vary at country-year, country-resource or countryresource-year level, is a country fixed effect, , is a year-resource fixed effect and . , is an error term. 9 The country fixed effects control for time-invariant country-specific characteristics such as geographic location and geological conditions. The year-resource fixed effects control for time-varying resource-specific factors such as international prices and technological progress. The additional controls included in , , are a measure of the previous stock of discoveries of resource in country as well as the square of this stock measure. This allows us to capture the country-specific dynamics related to the clustering of discoveries over time due to the probing effect and the depletion of geological reserves after large numbers of discoveries pushing extraction costs. 10 While the combination of country and year-by-resource fixed effects is our favoured specification, we also estimate regressions with year and country-resource fixed effects instead as this allows us to explicitly include natural resource prices (which we believe to yield a coefficient of interest in its own right) as a variable in , , . We cluster standard errors at the country-resource level but the significance of our results is unchanged if we cluster at the year-resource level (see Supplementary Appendix).

Identification
As discussed at length in the literature, resource discoveries can impact policies and institutions. For instance, discoveries may trigger conflicts over natural resources and erode political institutions (Ross 2001(Ross , 2012. As a first step to avoiding reverse causality, all explanatory variables are included with a lag. 11 Nevertheless, a naïve OLS estimation of (I) faces serious concerns. To try and isolate variation in openness which is exogenous to resource discoveries, and in the absence of a large-scale natural experiment, we require a strong instrument. Since both cross-sectional and time variation are important in our setting, we need the instrument to be time varying. This is not easy, since most previous instruments used for comparable variables in the literature tend to rely exclusively on cross-sectional variation. Our solution, which rests on a number of assumptions to be discussed below, is to construct an instrument for openness based on the idea that neighbours' market orientation, and in particular 9 The model is essentially a difference-in-difference specification where the key coefficient of interest is . 10 In those specifications where we include the past stock of discoveries (essentially the sum of lagged dependent variables) Nickel bias might be a concern (Nickell, 1981). However, given that it tends to 1/T and we have roughly 50 years of annual data, this bias is likely to be small. 11 The reverse causality generally discussed in the literature would tend to bias our results downwards -resourcerich countries are often shown to do worse on a number of institutional measures. where our hypothesis is that 0 and 0. For the instruments to be valid, they need to be strongly correlated with the openness of markets in country but uncorrelated with resource discoveries, conditional on openness of markets in country . The instruments satisfy the inclusion condition as we will show below. One concern with the exclusion restriction might be that discoveries in a neighbour (after that neighbour opens up) make exploration in country more attractive, independent of whether country also opens up (e.g., because of additional geological information). While this effect cannot be ruled out, it is likely to be at most a local effect. Information gained from a successful discovery only applies to a very limited geographical area, usually not more than several square kilometres. Nevertheless, we directly control for a distance-weighted measure of discoveries in other countries in the past year in a robustness exercise and find that our main empirical insights are unchanged.
12 Buera et al. (2011) include the lagged market orientation index in their reduced form specification which we exclude given the endogeneity concern. Their aim is not to construct an instrument, but to motivate a structural estimation. Furthermore, Buera et al. also include the distance-weighted growth rate of open countries, , | 1 ,as a complement to , | 0 . Including it does not change any results. But since the coefficient on , | 1 is never significant in the first stage we exclude it to have the most parsimonious instrument possible. 13 The weights are based on distance data obtained from the CEPII: http://www.cepii.fr/CEPII/fr/bdd_modele/presentation.asp?id=6 If both (regional) opening of markets and discoveries are driven by a (third) outside factor, omitted variables might also pose a threat to our identification. We try to address this in several ways. First, directly controlling for a lagged, distance-weighted measure of discoveries in other countries (as just discussed) addresses not only the regional spill-over problem but also the problem of omitted variables related to an unknown deep driver (such as US foreign policy perhaps or the fall of the Soviet Union) to some degree. Secondly, we show that the results are a general phenomenon, which holds for different time periods and regions -no one region or time period (which might be affected by a specific omitted variable) is responsible for the results. More generally, the rich fixed-effects structure we use, in particular the year-byresource effects, rule out a number of alternative stories such as one in which by coincidence the arrival of a new technology for discoveries of a particular resource happens concurrently with a wave of opening up of markets. The robustness section discusses this in more detail.

III. DATA AND STYLIZED FACTS
Here we discuss the various datasets that we use. We focus on the novel data on major hydrocarbon and mineral deposits as well as the data on market orientation (see Appendix B for a more comprehensive list of data and sources as well as additional summary tables).

Discoveries
Discoveries are our main dependent variable in our empirical model (I). The oil and gas discovery dataset is from Horn (2014). Horn reports discoveries of giant oil (including condensate) and gas fields which we refer to jointly as hydrocarbon or simply oil discoveries.
A giant discovery is defined as a discovery of an oil and/or gas field that contains at least 500 million barrels of ultimately recoverable oil equivalent. Ultimately recoverable reserves refer to the amount that is technically recoverable given existing technology.
The data on mineral deposits discoveries is from MinEx. The list of minerals included in the dataset is comprehensive and includes precious metals and rare earths. As in the case of hydrocarbons, we only capture discoveries above a certain threshold, corresponding roughly to a mineral deposit which has the capacity to generate an annual revenue stream of at least USD 50 million after accounting for fluctuations in commodity prices. The hydrocarbons and the mineral datasets themselves were constructed from many underlying sources. Minex constructed the data from company public reports (Annual Reports, press releases, NR 43-101 studies, etcetera), technical and trade journals (such as Economic Geology, Northern Miner and Mining Journal) and Government Files (from the various Geological Surveys). The data was up to date as of August 2013. Minex defines the discovery date of a mineral discovery as the moment when it was realized that the deposit has significant value, usually the date of the first economic drill intersection. Some deposits might have had small-scale operations in place prior to the discovery date -if there is an order-of-magnitude increase in the known size of the deposits, the date of the increase in the size of the discovery is taken as the discovery date.
Overall, we believe our data set to have the most comprehensive list of giant or big natural resource deposits for the period since 1950. Nevertheless, there are a number of constraints.
First, while only deposits above a certain threshold size are even considered for inclusion in the dataset (see the list in Appendix B), there is likely to still be a certain measurement bias towards larger deposits given better documentation for larger deposits. Secondly, data for certain countries such as Russia and China might be under-measured given less accessible data.
Last, our dataset excludes iron ore and bauxite. Since they are more abundant than other metals, there are not always well-timed discoveries to speak of and Minex excluded them from the dataset. Exploitation decisions tend to be based more on proximity to port facilities for iron ore and substantial energy availability for bauxite than other factors. Figure 2 shows a map with all natural resource discoveries included in our dataset. Figure 3 plots the total number of worldwide discoveries (split between minerals and hydrocarbons) by year. Since the early 1980s the average number of discoveries has been fairly stable at the global level. In our empirical work we use discoveries disaggregated by 128 countries and 33 types of resources.

Figure 2: Map of Worldwide Natural Resource Discoveries
Note: Based on data from MinEX for mineral discoveries and Mike Horn for hydrocarbon discoveries.

Figure 3: Number of Natural Resource Discoveries by Year
Note: Based on data from MinEX for mineral discoveries and Mike Horn for hydrocarbon discoveries.

Exploration
We will also use exploration effort, a key driver of discoveries, as dependent variable in our empirical model (I). To measure exploration effort, we use disaggregated data on exploration expenditures from Rystad for oil and gas and from SNL Metals and Mining for selected minerals including copper, nickel, zinc, diamonds, uranium, and platinum. The SNL Metals  1952 1954 1956 1958 1960 1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998

Measure of market orientation
Our prime explanatory variable in our empirical model (I) is market orientation or openness. For our purposes, we capture a broad measure of market orientation with policy implications which often reverberate on the "openness" of the resource sector. Indeed, investment in exploration is worthwhile only if there are prospects for further extractive activities. Such a generic measure of market orientation allows us to capture a combination of factors such as favourable business climate including fiscal terms, political risks and access to relevant equipment and financing. 15 We thus use the indicator as proxy for country's degree of market orientation.
It must be noted that a binary variable such as the SW indicator is necessarily restricted in how much it can measure. While we think of the SW indicator as proxy for the above list of factors, it is interesting to ask which of those which might be the most important ones in determining natural resource discoveries. Given that no other meaningful variable is available for most of the time period we are studying, we approach this question in section IV.D by investigating with which alternative measures of openness and institutions the SW index is most closely correlated.
A first look at the data: effect of opening markets on discoveries As a first look at the data, we conduct an event-study type of analysis where we calculate the average number of discoveries prior and after liberalization for all such episodes in the updated Wacziarg and Welch dataset. 16 Figure 4 shows that the number of discoveries significantly increases once economic liberalization has led to a more market-oriented economy. The average number of discoveries per year and country rises from 0.2 prior to liberalization to 0.36 afterwards. 17 The pattern linking liberalization and discoveries seems to hold across geographical regions and time periods. Looking at all episodes of market opening in our dataset (82 cases), we observe that in 54 percent of cases countries discovered more natural resources in the 10 years following a shift to a market-oriented economy than in the ten years before, in 24 percent of cases there was no difference, and in 22 percent of cases they discovered less.
difficult to reverse and thus more credible than say sectoral regulations that govern rather narrower aspects of a bilateral relationship between firms and national authorities. 16 In practice, we regress the number of discoveries on a set of period fixed effects while controlling for event fixed effects. We then retrieve the coefficients of the period fixed effects and plot them. 17 Discoveries seem to start increasing about 1-2 years prior to opening, perhaps due to an anticipation effect.

A. Benchmark Results
We now turn to our benchmark results. Table 2 gives the OLS estimates of the impact of openness on natural resource discoveries. In all specifications openness is estimated to have a highly significant positive impact on discoveries. Column 1 only includes year and countryby-resource fixed effects, as well as our basic variable of interest. In columns 2, 3 and 4 we add as additional controls the level of resource prices as well as the stock of past discoveries and its square. In column (5) we use country and year-by-resource fixed effects (and consequently we have to drop the price variable). 18 Given that price data is only available since the mid-1960s for many natural resources (and even later for some), we lose a fairly large number of observations in columns (2)-(4). 19 This, combined with the demanding nature of the year-by-resource fixed effects, makes column (5) our preferred specification of Table 2.  (1) and (5) are larger than in the other columns because columns (2) -(4) include resource prices as a control and price data are not available for certain years and resources. Column (5) uses country as well as year-resource fixed effects instead of the year and country-resource fixed effects employed in columns (1) -(4). *** denotes significance at the 1% level, ** at the 5% level, and * at the 10% level.
In terms of quantification, this specification also provides what we see as a safe lower bound -first it is a demanding specification and second, as we will see below, we believe the OLS estimates to be biased downwards given that the 2SLS estimates are substantially larger. The 18 For some resources, price data have limited time series information. Appendix B provides detailed information on available price data by natural resource. 19 In the shorter time period covered by regressions (ii)-(iv) the point estimates are significantly larger.
(1)   (1) and (5) are larger than in the other columns because columns (2) -(4) include natural resource prices as a control and price data are not available for certain years and resources. Column (5) uses country as well as year-resource fixed effects instead of the year and country-resource fixed effects employed in columns (1) -(4). The Anderson-Rubin Wald statistics allows for a test of the joint significance of the excluded instruments in a reduced-form estimation, where the null hypothesis is that the coefficients of the excluded instruments are jointly equal to zero. *** denotes significance at the 1% level, ** at the 5% level, and * at the 10% level.
We now turn to the instrumental variable results, which are based on the identification assumptions discussed previously. Table 3 shows the first-stage of the 2SLS estimates. The coefficients on distance-weighted average openness and on distance-weighted growth of closed economies have the expected sign. The former is always highly significant and positive while the latter is only significant in columns (1) and (5) which allow for longer timer series.
Overall, the high F-statistics (of the excluded instruments) indicate that the instruments are highly correlated with the endogenous variable.
(1) Year by Natural Resource Year by Natural Resource Year by Natural Resource Year by Natural Resource  (1) and (5) are larger than in the other columns, because columns (2) -(4) include natural resource prices as a control and price data are not available for certain years and resources. Column (5) uses country as well as year-resource fixed effects instead of the year and country-resource fixed effects employed in columns (1) -(4). *** denotes significance at the 1% level, ** at the 5% level, and * at the 10% level. OLS estimates are biased downwards in accordance with reverse causality (resource discoveries leading to rent seeking and potentially conflict).
While significance is not affected, the main coefficient of interest is significantly smaller in columns (1) and (5) than in columns (2), (3) and (4)  As discussed in section II, a more open market orientation (a lower implicit or virtual 'tax' on IRCs), but also higher prices and the stock of previous discoveries impact the number of new discoveries. Empirically, we find that increases in prices are significantly associated with more discoveries. The result is intuitive, since higher prices make additional exploration activity profitable. The coefficient associated with the stock of discoveries is positive and statistically significant. This suggests that in locations where discoveries have occurred in the past, more discoveries are more likely (an informational probing effect). 22 The coefficient associated with the square term in the stock of cumulative discoveries is negative suggesting that the effect is non-linear. In other words, bigger stocks of cumulative discoveries eventually turn out to be associated with a lower likelihood of discovery as the easiest available and cheapest deposits have been discovered ("running out of resources" effect). We interpret this as a trade-off between the initially reduced costs of exploring close to a known deposit with the eventually increased cost due to geological depletion.

B. Verifying the Mechanism: Exploration Efforts
So far, we have focused on the relationship between market orientation and major discoveries.
To examine the underlying mechanism, we explore whether exploration efforts rise following shifts in market orientation. Our hypothesis is that more market-oriented and open economies are able to attract more exploration investment and thus have more resource discoveries. Oil and gas exploration, as well as mineral exploration, are capital intensive and thus costly.
Nowadays, over a hundred billion dollars are spent on resource exploration annually according to Rystad and SNL Metals and Mining. And while exploration is a very risky activity 23 , in which "luck is obviously a major factor" (Harbaugh, Davis, and Wendebourg, 1995), exploration efforts ought to be a key determinant of discoveries. To verify that proposition for our data, we first estimate the following equation is a p-th order lag operator. 22 Cavalcanti et al (2016) used well-level data on oil drilling for Brazil to show that, after a first wild-cat discovery, follow-up exploration activity and additional discoveries increase significantly in following years. 23 An oil exploration well (wildcat well -a well drilled a mile or more from an area of existing oil production) can have a probability as low as 10% of yielding viable oil, while a rank wildcat (a well drilled in an area where there is no existing production) has an even smaller chance of finding oil. Elf was drilling in 1971 for offshore oil in Norway and found nothing. Recently, it found a huge new field just 3 metres away from the original drilling. Drilling outcomes are therefore highly uncertain. 24 Exploration expenditures data are deflated using the US GDP deflator. Using alternative deflators gives similar results.  Table 5 reports the results of this exercise. We always strongly reject the null hypothesis of no impact of exploration spending on discoveries at the 1% significance level.   (5) uses country as well as year-natural resource fixed effects instead of the year and countrynatural resource fixed effects employed in columns (1) -(4). *** denotes significance at the 1% level, ** at the 5% level, and * at the 10% level.
Having thus established that exploration spending increases the likelihood of discoveries, we test whether openness increases exploration spending to complete the causal chain. To do so, we estimate a regression analogous to equation (I). Table 6 gives the 2SLS estimates where we instrument openness as before. Unfortunately, the period of analysis when using exploration spending is relatively short due to exploration data only being available starting in 1994 (to the best of our knowledge it is nevertheless the best available data). Still, we find a strong positive impact of openness on exploration spending. The point estimates in Table 6 suggest an increase in exploration spending of close to 150% after opening markets when using the lowest point estimate (column 2). 25 This fits well with the above quantification of the impact of market orientation on the number of discoveries. Opening up the economy thus leads to a large and significant increase in exploration spending that in turn results in additional discoveries of new reserves according to Table 6. The level of prices is also found to be strongly positively associated with exploration spending, with an estimated elasticity of about one. Furthermore, the stock of cumulative discoveries first has a positive 'information' effect and eventually a negative 'depletion' effect as also found empirically for the effects on discoveries reported in Tables 2 and 4.

C. Robustness
We examine robustness by exploring a wide array of alternative specifications, additional controls, alternative definitions of the dependent variable, countries and periods excluded from our sample, splitting up the sample between hydrocarbon and mineral discoveries, using an alternative estimator which specifically takes into account the large number of zeros in the discovery data, and collapsing our data to a two-way, country-year panel (see Tables S1-S11 in the Supplementary Appendix). As discussed in section II, in theory it might be true that the 25 This can be calculated as 100 * 1 which with  = 0.903 gives 147%. exclusion restriction for the distance-weighted instrument is not satisfied as openness of neighbours' economies increases discoveries there and in the home country as the neighbours' discoveries provide additional geological information about the home country. By controlling directly for neighbours' distance-weighted discoveries, this concern is addressed (see Table   S1). We find that the point estimate on market orientation is virtually unchanged relative to the baseline regression, but the estimate for the impact of neighbours' discoveries is always positive albeit not always significant.
The significance of our core results in Table 4 is unaffected if standard errors are clustered at the country-resource level instead of at the year-resource level (see Table S2). If the human capital index and GDP from the Penn World Tables as are added as controls in our 3-way panel, the results are broadly the same even though the point estimates of the effect of openness on discoveries are somewhat smaller (see Table S3).
To allow for a more even comparison between countries (even though the fixed effects included in all regressions already address issues such as different sizes of countries), we also estimated the regressions with discoveries per capita as the dependent variable (see Table S4).
The point estimate on openness remains positive and highly significant. With discoveries per capita as the dependent variable the estimates are in fact particularly robust and are both qualitatively and quantitatively unchanged when additional control variables (such as GDP per capita) are added (Table S5). Instead of using a count variable (the number of discoveries), we have also used a simple dummy variable as the dependent variable, again confirming the significance of market orientation for natural resource discoveries (see Table S6).
Our core empirical results also hold when excluding any of the individual country groups (see Table S7) or excluding any particular decade (see Table S8). Across all regressions the coefficient on market orientation remains positive and significant. This is important since it suggests that our results are not driven by one region or one specific time period only.
Given our rich data set for discoveries, we can estimate equation (I) individually for different   natural resources ∈  ,  ,  ,  ,  ,  , , . This list covers over 90% of all discoveries for our period of analysis, 1950-2004, corresponding to 2171 out of a total of 2392 discoveries. The estimates for the impact of openness on discoveries range from virtually no impact for uranium to a (statistically significant) increase of close to 3% for nickel and oil and an increase of over 5% for silver, albeit that the latter is not precisely estimated (see Table S9). 26 The results indicate that investments (and thus discoveries) of different natural resources are potentially differentially sensitive to a country's institutional environment.
Table S10 employs the zero-augmented Poisson estimator (ZIP) as an alternative estimator.
This allow us to directly model the fact that our dependent variable is count data with a very large fraction of zeros. In particular, ZIP fits a logit model to predict the excess zeros and separately models the count data by fitting a Poisson model. 27 To predict excess zeros we use the lag of the previous stock of discoveries. 28 The coefficient for the effect of openness on discoveries is positive and significant. The interpretation of the point estimate is now different.
For example, opening up increases the expected log count of discoveries by 0.486 (from the coefficient in column 4 of Table S10) so that discoveries increase by a factor of . ~1.7 after a country opens up. This is somewhat larger than the quantitative effect obtained from our OLS or 2SLS estimates in Tables 2 and 4.
As an additional exercise, we collapsed our three-way panel (country, year, resource) to a twoway panel (country, year) since the obtained regression coefficients are particularly easy to interpret and can be immediately compared to the preliminary event-study analysis conducted in section III. Opening of the economy increases discoveries by 0.47 per year and country (column 1 of Table S11). Recall that a cursory look at the data as shown in Figure 4 suggested an increase of 0.16, severely underestimating the positive impact.
26 These numbers are obtained by taking the estimated coefficients for each resource k and dividing them by the average yearly number of discoveries of resource . The drawbacks of ZIP are that we do not employ an IV strategy and for computational reasons we cannot include a fixed-effects structure which is as rich as in the least squares estimations.
Last, one could argue that there are important differences in the role market orientation plays in fostering mineral versus hydrocarbon discoveries. In particular, minerals might be seen as more appropriable than hydrocarbons because mining output does not move through pipelines and takes place exclusively onshore. Instead our results suggest that in fact the effect of market orientation is driven as much by hydrocarbon as mineral discoveries (columns 2 and 3 of Table   S11). 29

D. A look at what the SW Openness Indicator might be proxying?
As explained in the data section, market orientation is defined as the absence of the following features: (i) average tariff rate on imports above 40%; (ii) non-tariff barriers covering more than 40% of imports; (iii) a socialist economy; (iv) the state holds a monopoly of the major exports; and (v) a black market premium above 20%. In essence, countries defined as market oriented thus allow for relatively free trade and do not exercise state control over the main exports. One might see it as quite intuitive that such a situation should encourage more exploration investment and thus discoveries than the contrary one. One might also wonder, however, what more detailed features of the economy market orientation might be proxying for. To get a sense of the effects of unbundling market orientation, Table 7 shows partial correlations between the SW/Wacziarg indicator and various International Country Risk Guide (ICRG) Political Risk Rating sub-indices which are available since 1984.

Table 7: Partial correlations of SW/Wacziarg Openness Indicator with ICRG Indices
Note: The table reports partial correlation coefficients between the SW/Wacziarg Openness indicator and ICRG political risk sub-indices. The overlap between the data series is 1984-2004. 29 We lose some power when splitting the sample, hence the reduced significance levels.  The most correlated ICRG component is the investment profile index -a direct measure of how attractive it is to invest in a certain country. The index measures contract viability/expropriation, profit repatriation and payment delays. The index goes from 0-12 with a higher score indicating less risk. The average value for Latin America and Sub-Saharan Africa increased from 5 to 7.5 between 1984 and the 2000s, in line with the increase in discoveries observed in these regions. In summary, while we are not able to fully disentangle the channels which link market orientation and natural resource discoveries, it seems that the investment climate -proxied by property rights and payment security -is one of the key factors which the SW index proxies for.

V. CONCLUSION
We have examined the effect of changes in market orientation on proven (known) natural wealth. Consistent with the predictions of a two-region model, we presented empirical estimates based on a large panel of countries that show that increased market orientation causes a significant increase in discoveries. In a thought experiment whereby economies in Latin America and sub-Saharan Africa remained closed, they would have only achieved one quarter of the actual increase in discoveries they have experienced since the early 1990s. These results help explain the worldwide shift in the geographic distribution of natural resource discoveries, with the economic opening in Sub-Saharan Africa and Latin America contributing to a large increase in the share of worldwide discoveries in these two regions.
Our results provide novel evidence in support of the primacy of outward market orientation by calling into question the view that natural resource endowments are an exogenous feature of an economy. And while we find that market orientation increases discoveries, they do not rule out that the resulting resource boom may lead to aggressive rent seeking and conflicts, potentially worsening welfare for citizens. Our results thus suggest that the relationship between resource endowments and institutions (in the larger sense) is complex. (1 ) .
Maximizing net worth also gives the efficiency conditions that yield initial reserves and exploration in much the same way as discoveries follow from (A3):

p G S I G S I R S I r p G S I G S I R S I r T
The main difference is that initial exploration investment benefits from making future extraction cheaper by ensuring higher proved reserves. We thus get the following comparative statics results for discoveries, exploration investment and reserves: where the pluses and minuses indicate the signs of the partial derivatives.
Finally, to explain world resource prices one needs to introduce global resource demand. Let world demand for oil in period t be iso-elastic and given by and .
Using initial exploration and discoveries as given in (4), the depletion equations become (A7)  , .   One could also extend the analysis to allow for discoveries to depend on how much has been explored initially and thereby on geological conditions. One can capture this by making A a function of 0

I and *
A a function of * 0 , I but our main conclusions regarding the shifting frontier of natural resources will not be materially affected. Overall, discoveries of Gold, Oil, Natural Gas, Copper, Nickel, Uranium, Zinc and Silver account for over 90 percent of all discoveries (followed by Diamonds and Molybdenum which account for 1-2 percent each).   Chromium (1990), Copper, Diamonds (2005, Fluorite (no price data), Natural Gas (1985), Gold, Graphite (no price data), Lead, Lithium (1997), Magnesium (2003, Manganese, Mineral sands (no price data), Molybdenum (2012), Nickel, Niobium (2013), Oil, Palladium (1992, PGE (no price data), Phosphate, Platinum (1992), Potash (no price data), Rare earths (no price data), Silver (1968), Soda ash (2007), Tantalum (2009), Tellurium (2013), Tin, Tungsten, Uranium (1980), Vanadium (1987), Zinc, Zircon (1997.  1990-1999 2000-2009 1950-1959 1960-1969 1970-1979 1980-1989 SUPPLEMENTARY APPENDIX Table S.1 adds neighbours' distance-weighted discoveries as an additional control and finds that the point estimate on market orientation is virtually unchanged relative to the baseline regression whilst the estimate for the impact of neighbours' discoveries is always positive but not always significant.  (1) and (5) are larger than in the other columns because columns (2) -(4) include resource prices as a control and price data are not available for certain years and resources. Column (5) uses country as well as year-resource fixed effects instead of the year and country-resource fixed effects employed in columns (1) -(4). *** denotes significance at the 1% level, ** at the 5% level, and * at the 10% level.

APPENDIX B: DATA
(1)    (1) and (5) are larger than in the other columns because columns (2) -(4) include resource prices as a control and price data are not available for certain years and resources. Column (5) uses country as well as year-resource fixed effects instead of the year and country-resource fixed effects employed in columns (1) -(4). *** denotes significance at the 1% level, ** at the 5% level, and * at the 10% level.
(1)      country, year and resource on countries' SW/Wacziarg openness indicator as well as a number of controls. Distance-weighted average openness and distance-weighted growth of closed economies are used as excluded instruments for the SW/Wacziarg Openness indicator. *** denotes significance at the 1% level, ** at the 5% level, and * at the 10% level.
The effect of market orientation on natural resource discoveries remains positive and highly significant.  Table S7 shows how remarkably stable our results are to excluding individual country groups as whatever group is removed from the sample the coefficient on openness remains positive and significant. Note: The table reports the second stage of a 2SLS estimation regressing the number of resource discoveries by country, year and resource on countries' SW/Wacziarg openness indicator as well as a number of controls. Distance-weighted average openness and distance-weighted growth of closed economies are used as excluded instruments for the SW/Wacziarg Openness indicator. Each column excludes one specific geographic area which is specified in the "Excluded Region" row. *** denotes significance at the 1% level, ** at the 5% level, and * at the 10% level.
(1)  Table S8 shows that the results are robust to excluding individual time periods. Again, across all regressions the coefficient on openness remains positive and significant. Note: The table reports the second stage of a 2SLS estimation regressing the number of resource discoveries by country, year and resource on countries' SW/Wacziarg openness indicator as well as a number of controls. Distance-weighted average openness and distance-weighted growth of closed economies are used as excluded instruments for the SW/Wacziarg Openness indicator. Each column excludes one decade which is specified in the "Excluded Decade" row. *** denotes significance at the 1% level, ** at the 5% level, and * at the 10% level.

Table S9: The Impact of Liberalization on Resource Discoveries (2SLS) -Results by natural resource
Note: The table reports the second stage of a 2SLS estimation regressing the number of resource discoveries by country and year on countries' SW/Wacziarg openness indicator as well as a number of controls. Distance-weighted average openness and distance-weighted growth of closed economies are used as excluded instruments for the SW/Wacziarg Openness indicator. Each column excludes shows results for one main type of natural resource. *** denotes significance at the 1% level, ** at the 5% level, and * at the 10% level.
(1)  Table S10 present estimates analogous to those in table 3 with the zero-augmented Poisson estimator (ZIP) to allow for count data with a very large fraction of zeros. To predict excess zeros we use the lag of the previous stock of discoveries. The coefficient for the effect of openness on discoveries is again positive and significant in all specifications. Note that the coefficient is not directly comparable to the OLS/2SLS ones given the Poisson regression.   ). Opening of the economy increases discoveries by 0.47 per year and country (column 1). Splitting the analysis between hydrocarbon and mineral deposits shows that the effect of openness on discoveries remains positive and statistically significant (columns 2 and 3). Note: The table reports the second stage of a 2SLS estimation regressing the number of resource discoveries by country and on countries' SW/Wacziarg openness indicator as well as a number of controls. Distance-weighted average openness and distance-weighted growth of closed economies are used as excluded instruments for the SW/Wacziarg Openness indicator. *** denotes significance at the 1% level, ** at the 5% level, and * at the 10% level. (1)