Evidence from a Natural Experiment on the Development Impact of Windfall Gains: The Camisea Fund in Peru

Corral, Leonardo; Henderson, Heath; Miranda, Juan Jose Working Paper Evidence from a Natural Experiment on the Development Impact of Windfall Gains: The Camisea Fund in Peru IDB Working Paper Series, No. IDB-WP-687 Provided in Cooperation with: Inter-American Development Bank (IDB), Washington, DC Suggested Citation: Corral, Leonardo; Henderson, Heath; Miranda, Juan Jose (2016) : Evidence from a Natural Experiment on the Development Impact of Windfall Gains: The Camisea Fund in Peru, IDB Working Paper Series, No. IDB-WP-687, Inter-American Development Bank (IDB), Washington, DC, https://hdl.handle.net/11319/7520 This Version is available at: https://hdl.handle.net/10419/146470 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. http://creativecommons.org/licenses/by-nc-nd/3.0/igo/legalcode Evidence from a Natural Experiment on the Development Impact of Windfall Gains: The Camisea Fund in Peru Leonardo Corral Heath Henderson Juan Jose Miranda IDB WORKING PAPER SERIES Nº IDB-WP-687 March 2016 Office of Strategic Planning and Development Effectiveness Inter-American Development Bank March 2016 Evidence from a Natural Experiment on the Development Impact of Windfall Gains: The Camisea Fund in Peru Leonardo Corral Heath Henderson Juan Jose Miranda Inter-American Development Bank Drake University The World Bank Cataloging-in-Publication data provided by the Inter-American Development Bank Felipe Herrera Library Corral, Leonardo. Evidence from a natural experiment on the development impact of windfall gains: the Camisea Fund in Peru / Leonardo Corral, Heath Henderson, Juan José Miranda. p. cm. — (IDB Working Paper Series ; 687) Includes bibliographic references. 1. Intergovernmental fiscal relations-Peru. 2. Oil and gas leases-Peru. 3. Municipal budgets-Peru. 4. Natural gas-Peru. 5. Economic development-Peru. I. Henderson, Heath. II. Miranda, Juan José. III. Inter-American Development Bank. Strategy Development Division. IV. Title. V. Series. IDB-WP-687 OPTIONAL: Type address for correspondence OPTIONAL: Type Authors name and eMail Copyright © Inter-American Development Bank. This work is licensed under a Creative Commons IGO 3.0 AttributionNonCommercial-NoDerivatives (CC-IGO BY-NC-ND 3.0 IGO) license (http://creativecommons.org/licenses/by-nc-nd/3.0/igo/ legalcode) and may be reproduced with attribution to the IDB and for any non-commercial purpose, as provided below. No derivative work is allowed. Any dispute related to the use of the works of the IDB that cannot be settled amicably shall be submitted to arbitration pursuant to the UNCITRAL rules. The use of the IDB's name for any purpose other than for attribution, and the use of IDB's logo shall be subject to a separate written license agreement between the IDB and the user and is not authorized as part of this CC-IGO license. Following a peer review process, and with previous written consent by the Inter-American Development Bank (IDB), a revised version of this work may also be reproduced in any academic journal, including those indexed by the American Economic Association's EconLit, provided that the IDB is credited and that the author(s) receive no income from the publication. Therefore, the restriction to receive income from such publication shall only extend to the publication's author(s). With regard to such restriction, in case of any inconsistency between the Creative Commons IGO 3.0 Attribution-NonCommercial-NoDerivatives license and these statements, the latter shall prevail. Note that link provided above includes additional terms and conditions of the license. The opinions expressed in this publication are those of the authors and do not necessarily reflect the views of the Inter-American Development Bank, its Board of Directors, or the countries they represent. http://www.iadb.org 2016 Evidence from a Natural Experiment on the Development Impact of Windfall Gains: The Camisea Fund in Peru∗ Leonardo Corral Office of Strategic Planning and Development Effectiveness Inter-American Development Bank Heath Henderson† College of Business and Public Administration Drake University Juan Jose Miranda Environment and Natural Resources Global Practice The World Bank February 24, 2016 Abstract We study the economic effect of windfall gains by examining a Peruvian natural experiment. The Camisea Fund for Socioeconomic Development (FOCAM) is an inter-governmental fiscal transfer scheme that allocates natural gas royalties generated by the Camisea Gas Project to eligible subnational governments. We exploit the rules governing FOCAM allocation to identify the effect of the transfers on municipal accounts, local infrastructure, and economic development. Using a newly constructed districtlevel dataset for the years 2005 and 2012, we find evidence of positive impacts on municipal capital expenditures and local infrastructure. However, we also find evidence of a negative impact on municipal current expenditures. More specifically, we find that municipalities with low absorptive capacity coped with the increased administrative burden of FOCAM transfers by reallocating administrative effort toward (away from) executing capital (current) expenditures. Key words: Inter-governmental Fiscal Transfers; Natural Experiment; Peru; Windfall Gains JEL codes: C14; H70; O10; O54; Q32 ∗The authors would like to thank Edwin Go˜ni and Brigitt Bencich Aguilar for supporting data collection efforts, Naijun Zhou for processing the nighttime lights data used in the analysis, Elizabeth Brito for providing key information and useful suggestions for the write up of the Camisea project, and the Inter-American Development Bank (IDB) for financial support. Further, for their helpful comments we would like to thank participants of SPD’s “Half Baked Lunch” seminar series. The views expressed here are those of the authors and should not be attributed to the IDB or its member countries. †Corresponding author; 359 Aliber Hall, Des Moines, IA 50311; Tel: +1 515 271 2898; Email: [email protected]. 1 Introduction The economic effect of windfall gains (e.g. foreign aid, natural resource rents, and inter-governmental fiscal transfers) remains controversial, due in part to identification issues in the empirical literature.1Consider, for example, the empirical literature on the effect of foreign aid on economic growth. Using 2SLS in the context of cross-country growth regressions, Burnside and Dollar (2000) found that aid only positively affected growth in developing countries with sound fiscal, monetary, and trade policies. Hansen and Tarp (2001), however, found an unconditionally positive aid-growth relationship when additionally accounting for endogeneity bias due to country-specific effects. Rajan and Subramanian (2008) questioned the instrumentation strategies used in prior cross-country studies and alternatively modeled the supply of aid based on donor (as opposed to recipient) characteristics. Contrary to the aforementioned studies, Rajan and Subramanian found little evidence of an aid-growth relationship, but their instrumentation strategy itself was questioned by Bazzi and Clemens (2013).2 The empirical literature on the economic effect of natural resource rents has also encountered identification issues. Also using cross-country growth regressions, Sachs and Warner (1995) found an inverse relationship between economic growth and the (initial) ratio of natural resource exports to GDP. Brunnschweiler and Bulte (2008), however, argued that Sachs and Warner’s resource dependence measure was endogenous to underlying structural factors and became statistically insignificant when instrumented. The authors then contended that a measure of resource abundance (i.e. subsoil assets) was potentially exogenous, and found a positive relationship between resource abundance and economic growth in a cross-country setting. Van der Ploeg and Poelhekke (2010) nevertheless suggested that Brunnschweiler and Bulte’s measure of subsoil assets was itself endogenous given its proportionality to resource rents. The authors then found that the subsoil assets variable became statistically insignificant when instrumented, though they cautioned that their instrument was also “somewhat endogenous” (pg. 47). In light of such identification issues, van der Ploeg (2011) suggested that “[t]he road forward might be to exploit variation within a country where . . . the danger of spurious correlation is minimized” (pg. 381). While research at the subnational (or regional) level may offer additional insights into the economic effect of foreign aid or natural resource rents, it is also relevant to understanding another increasingly important 1We follow Dalgaard and Olsson (2008) and define windfall gains by their disproportionate revenue-to-cost ratio as compared to revenues from the standard production of goods and services. See Arndt et al. (2010) for a review of the foreign aid literature and van der Ploeg (2011) for a review of the literature on natural resource rents. 2Bazzi and Clemens argued that Rajan and Subramanian were effectively instrumenting foreign aid with recipient population size, even though Rajan and Subramanian themselves stated that such an instrument was unlikely to satisfy the exclusion restriction. 1 type of windfall gain: inter-governmental fiscal transfers. Recent decades have seen a movement toward fiscal decentralization in many transition and developing economies (De Mello 2000; Arzaghi and Henderson 2005).3As the objective of decentralization is typically to improve allocative efficiency, distributional equity, or macroeconomic growth/stability, complex systems of inter-governmental transfers commonly accompany decentralization efforts (Bird 1993; Bird and Smart 2002). Subnational governments in developing countries are highly and increasingly dependent on these transfers (Gadenne and Singhal 2013),4but relatively little is known about their economic effect (Paler 2011; Becker et al. 2013). Paler (2011) discussed the theoretical mechanisms through which windfall gains can affect economic outcomes at the subnational level. First, Dutch disease can operate at the subnational level if the inflow of capital induces an increase in the local price level. While the windfall can provide local governments with revenue to fund long-run development, rising factor prices can also squeeze segments of the tradable-goods sector and hamper productivity growth. Second, windfall gains can strengthen or undermine local governing institutions. Strengthening can occur when additional resources facilitate capacity building and undermining can occur when windfalls induce corruption or rent-seeking behavior. Finally, windfall gains can hamper economic growth by inciting local conflict, particularly by exacerbating grievances due to social inequality, forced migration and environmental degradation, or ethnic tensions. While it is evident that the economic effect of subnational windfall gains is theoretically indeterminate, the empirical evidence also remains inconclusive. Caselli and Michaels (2009), for example, exploited variation in oil output across Brazilian municipalities to examine the economic effect of natural resource windfalls. The authors found little evidence of Dutch disease-type effects or improvements in local living standards, even though oil abundance caused municipal revenues and spending to increase. Further examining Brazilian municipalities, Brollo et al. (2010) used regression discontinuity design to analyze the effect of federal transfers on political corruption. Consistent with anecdotal evidence discussed in Caselli and Michaels, the authors found that larger transfers increased political corruption and reduced the quality of candidates for mayor.5 While the Brazilian evidence suggests that windfall gains can undermine local governing institutions, positive effects have been witnessed in other contexts. Becker et al. (2013), for example, analyzed the economic effect of a regional transfer scheme in the European Union. Also using regression discontinuity design, the authors found that those regions with 3Dillinger (1994) claimed that 63 out of the 75 developing and transition economies with populations greater than 5 million people have embarked or are embarking on some form of fiscal or political decentralization. 4More specifically, Gadenne and Singhal (2013) document that the average share of non-tax revenues in subnational revenues (i.e. the average fiscal gap) increased in developing countries from 49 percent to 62 percent between 1996-2000 and 2006-2010. 5See Vicente (2010) for further evidence suggesting that natural resource rents tend to induce corruption. 2 sufficient human capital and relatively high-quality institutions witnessed faster per capita income growth. Exploiting an exogenous upsurge in coca prices and cultivation in Colombia, Angrist and Kugler (2008) found some positive economic effects (e.g. increased self-employment earnings) of windfall gains as well. While the authors further witnessed increased violence in regions where coca cultivation increased, Dube and Vargas (2013) found that price increases for other Colombian exports (e.g. coffee) can mitigate conflict by reducing labor supplied to violent resource appropriation. Given the evident lack of resolution in the existing research, we contend that further empirical work is needed to better understand the effect of windfall gains. We thus examine the economic effect of windfall gains by exploiting a Peruvian natural experiment. Operational in 2004, the Camisea Gas Project – a natural gas extraction and distribution project – is one of Peru’s largest energy infrastructure projects. Largely through its two natural gas pipelines, the project impacts six of Peru’s 25 regions as well as the province of Lima. To support the economic, social, and environmental development of the regions affected by the pipelines, the Peruvian government established the Camisea Fund for Socioeconomic Development (FOCAM). FOCAM is an inter-governmental fiscal transfer scheme that allocates a percentage of the natural gas royalties received by the central government to eligible subnational governments. We exploit the rules governing FOCAM allocation to identify the effect of the transfers on municipal accounts, local infrastructure, and economic development. Using a newly constructed district-level dataset for the years 2005 and 2012, we first find that treated municipalities witnessed statistically significant increases in capital expenditures. Given that FOCAM transfers are earmarked for capital expenditures, this result is expected. Second, we find that treated districts also witnessed statistically significant positive impacts on local infrastructure. For example, we find that treated districts had greater access to internet and constructed more primary roads than their control group counterparts. Finally, we unexpectedly find evidence of a negative impact on municipal current expenditures. Anecdotal and quantitative evidence suggests that capacity constraints (e.g. lack of qualified technical staff) have limited budget execution rates, and we find that low capacity municipalities coped with the increased administrative burden of FOCAM transfers by reallocating administrative effort toward (away from) executing capital (current) expenditures. In what follows, Section 2 provides background information on the Camisea Gas Project and the associated FOCAM transfers. Section 3 discusses the available data, Section 4 outlines our identification strategy, and Section 5 presents the results of our econometric analysis. Finally, Section 6 provides discussion and concluding remarks. 3 2 The Camisea Gas Project In this section, we first provide a brief overview of Peru’s recent decentralization efforts so as to contextualize the Camisea Gas Project. We then turn to discussing the Camisea Gas Project in detail, after which we describe the rules governing allocation of the Camisea Fund for Socioeconomic Development (FOCAM). Finally, we conclude the section by arguing that receipt of FOCAM transfers is exogenous to select regions. Approval and promulgation of the Constitutional Reform Law of 2002 launched the decentralization process in Peru. The reform established that the territory of Peru was to consist of regions, provinces, and districts. Specifically, the country was divided into 26 units: 25 regions and the province of Lima. The regions were subdivided into provinces, and all provinces were subdivided into districts. In each jurisdiction, national, regional, and local governments were constituted and organized. The constitutional reform added that the sphere of regional governments was the regions and that the sphere of local governments was the provinces, districts, and towns. Several further legislative acts were also approved to promote administrative and fiscal decentralization, and to enhance citizen participation, particularly in yearly budgeting exercises. In addition, in order to optimize the use of public resources for investment, the National System for Public Investment (SNIP) was extended to regulate all regional and local governments (Contralor´ıa General de la Rep´ublica del Per´u 2014). Loayza et al. (2014) documented that these rapid changes overwhelmed many municipalities. On the one hand, decentralization transferred additional responsibilities and resources from the central government to regional and local governments. On the other hand, strong fiduciary requirements and demanding budgeting guidelines made budgeting and execution more difficult. Many municipalities have thus struggled to spend their allocated budget. In 2009, for example, municipalities spent on average 74 percent of their budgeted expenditures. Further, current expenditures tended to be executed at better rates than capital expenditures (83 percent versus 71 percent, respectively). Anecdotal and quantitative evidence suggests that capacity constraints (e.g. lack of qualified technical staff) have limited execution rates considerably, particularly for capital expenditures (see Loayza et al. [2014] for further information). It is in this context that the Camisea Gas Project was implemented. The Camisea Gas Project is Peru’s biggest energy project. Located in the department of Cusco, the Camisea natural gas fields have proven reserves of 9 trillion cubic feet, among the largest in Latin America. Royal Dutch/Shell started exploring the fields in the mid-1980s, but walked away in 1998 after contractual disputes with the Government of Peru (GoP) (The Economist 2003). Peru’s Private Investment Commission (COPRI) then issued an international call for proposals to further develop the fields. The Camisea Gas 4 where the first term is observable as Y1is observable for the treated districts and the second term is unobservable as Y0is not observed for the treated districts. It is thus impossible to directly observe the causal effect and this is the “fundamental problem of causal inference” (Holland 1986). Consider then an estimator of τ– denoted by ˆτ– as the coefficient on Tfrom a regression of Yon a constant and T. King and Zeng (2006) derived the following decomposition of the bias of ˆτas an estimator of τ: (2) E(ˆτ−τ)=∆o+ ∆p+ ∆i+ ∆e where ∆odenotes omitted variable bias, ∆pdenotes post-treatment bias, ∆idenotes interpolation bias, and ∆edenotes extrapolation bias. Omitted variable bias, ∆o, occurs when variables that are correlated with both treatment status and the dependent variable are omitted from the regression. Post-treatment bias, ∆p, is the result of controlling for variables that are themselves a consequence of treatment. Interpolation bias, ∆i, is due to a failure to properly adjust for independent variables within the observed range of the data. Finally, extrapolation bias, ∆e, results when there is a failure to properly adjust for independent variables when extrapolating beyond the observed range of the data. The above bias decomposition identifies the potential issues associated with estimating τusing observational data and permits us to explicitly address the assumptions that accompany our identification strategy. To this end, let Xdenote a vector of control variables, the appropriate choice of which enables us to reasonably make three simplifying assumptions. First, King and Zeng (2006) demonstrate that if (3) E(Y0|T= 1, X) = E(Y0|T= 0, X) holds, then the first component of Eq. (2) vanishes (i.e. ∆o= 0).18 Given the natural experiment (see Section 2) and a rich set of control variables (see Section 3), we contend that assuming ∆o= 0 is indeed reasonable. Second, if those characteristics included in Xinclude only pre-treatment characteristics (i.e. characteristics that are not a consequence of treatment status), then we may also reasonably assume ∆p= 0. Finally, as interpolation bias arises when controlling for Xusing the wrong functional form (e.g. linear rather than quadratic), with the appropriate regression diagnostics we may also assume ∆i= 0. 18That is, if the appropriate set of control variables is included, treatment assignment is random and omitted variable bias is eliminated. 11 With the above assumptions, Eq. (2) becomes E(ˆτ−τ)=∆eand we now turn to our strategy to mitigate this extrapolation bias. Extrapolation bias occurs when some members of the treated (control) group witness certain values of Xwith a positive probability that no members of the control (treated) group witness. Matching methods are data pre-processing algorithms that can serve to mitigate such extrapolation bias by discarding observations outside the region of common support. To this end, we use a matching method called “Coarsened Exact Matching” (CEM) (Iacus et al. 2011, 2012), as it possesses a number of desirable statistical properties that other matching methods (e.g. propensity score matching) do not possess. First, CEM is a “Monotonic Imbalance Bounding” (MIB) estimator, which means that the maximum degree to which a variable can be out of balance is pre-specified. Second, CEM meets the “congruence principle,” which states that the data space and the analysis space should be identical. Finally, and critically for our purposes, CEM automatically restricts the data to common support. The CEM algorithm proceeds in four stages. First, we create a copy X∗of the covariates X. Second, we coarsen X∗by pre-specified cutoff point or binning algorithm. Third, for each unique observation of X∗we create one stratum and place the observations in those strata. Finally, we drop any observation that does not belong to a stratum containing at least one treated and one control unit, and then assign the remaining strata to the original data X(Blackwell et al. 2009). After pruning observations, we can then conduct the analysis by regressing Yon Tand X, though when different numbers of treated and control units appear in different strata these regressions must be appropriately weighted (Iacus et al. 2012). Finally, it is important to note that the procedure can prune treated units thus changing the estimand to a “local average treatment on the treated.” Coarsening is central to CEM and, while our discrete covariates do not require further coarsening, it is necessary for the continuous covariates. Neither substantive knowledge of these continuous covariates nor inspection of the data reveals any obvious cutoff points. We thus use automated coarsening and employ two alternative approaches to assess the sensitivity of our results to the coarsening algorithm. Following Iacus et al. (2012), the first approach uses a leading binning rule developed by Shimazaki and Shinomoto (2007). As this approach imposes uniform bin sizes, our second approach uses the optimal k-means clustering algorithm described in Wang and Song (2011).19 Both of these procedures yield coarsened data that results in too few matched units and, as such, we couple each approach with the inductive relaxation procedure described in Iacus et al. (2012). This relaxation procedure seeks to increase the number of matched units 19k-means clustering is a popular approach to cluster analysis. While standard k-means algorithms do not guarantee optimality, the Wang and Song (2011) algorithm is optimal for one-dimensional clustering. The number of clusters for each variable in this procedure is determined by the Bayesian information criterion. 12 by strategically reducing the number of bins or clusters for each variable. More specifically, the procedure discriminates between alternative relaxations by minimizing a multivariate imbalance measure.20 With the appropriate coarsening choices, we can then reasonably assume E(ˆτ−τ) = 0 (i.e. ∆e= 0 and our estimator is unbiased) and turn to discussing our empirical model.21 The empirical model is as follows: (4) yi,2012 =α+τTi+βXi,2005 +εi where, for district i= 1,2, . . . , N at time t∈ {2005,2012},ydenotes the dependent variable, Tis the treatment indicator, Xdenotes a vector of independent variables (see Table 1 for a complete listing), and εis the error term.22 We estimate the model for all dependent variables (see Table 1) using the appropriate linear or non-linear model.23 For each dependent variable, the model is estimated (1) without additional controls or matching; (2) with controls and without matching; (3) with controls and matching using ShimazakiShinomoto coarsening; and (4) with controls and matching using k-means coarsening. Finally, as the sphere of influence of the local governments is the province, all standard errors are clustered at the province level. As will be seen, we find evidence that receipt of FOCAM transfers reduced municipal current expenditures (current). Following the discussion in Section 2, we hypothesize that municipalities with low absorptive capacity coped with the increased administrative burden of FOCAM transfers by reallocating administrative effort toward (away from) executing capital (current) expenditures. “An issue repeatedly highlighted in our interviews with municipal managers is the lack of necessary personnel . . . municipalities that enjoy a larger professional staff obtain higher execution rates” (Loayza et al. 2014, pg. 65). We thus proxy municipal absorptive capacity with (the log of) the number white-collar staff per 10,000 people (staff ). To test our hypothesis, we simply interact staff with the treatment indicator in our model for current expenditures. We then estimate the model using OLS on three alternative samples: (1) without matching; (2) with matching using Shimazaki-Shinomoto coarsening; and (3) with matching using k-means coarsening. 20While Iacus et al. (2012) use a variable-by-variable approach to sequentially relax coarsening choices, we find that further gains can be made by considering combinations of relaxations. We thus randomly sample with 10,000 draws the entire space of feasible relaxations, though our solution concept remains the same as in Iacus et al. (2012). Computer code is available upon request. 21Importantly, we can assess the reduction in covariate imbalance by examining post-matching normalized differences (Imbens and Wooldridge 2009). 22To mitigate post-treatment bias, note that we only control for pre-treatment characteristics. 23That is, OLS is used for continuous outcome variables, logit models are used for binary outcomes, Poisson or negative binomial regressions are used for count data, and Tobit models are used for censored outcomes. The estimated treatment effect is then calculated as the marginal effect (evaluated at the mean) of Tin each model. 13 5 Results The results of our analysis are presented in Tables 2-4 and Figure 4. The final two columns of Table 2 present the post-matching normalized differences for all independent variables. The ND-SS (ND-KM) column presents normalized differences calculated after using CEM with the Shimazaki-Shinomoto (k-means) algorithm to coarsen continuous covariates. First, note the sample size (N). It is evident that the matching procedure reduces the number of observations from 1,492 to 1,095 and 1,107 for the Shimazaki-Shinomoto and k-means approaches, respectively. Second, relative to the pre-matching normalized differences (NDPM), we see that the ND-SS and ND-KM columns present substantially reduced normalized differences. For example, the normalized difference for literacy is reduced from -0.29 to -0.02 and -0.03 in the ND-SS and ND-KM columns, respectively. Finally, for both approaches and across all covariates, we see that there remains no statistically significant differences in means after CEM. CEM thus substantially reduces covariate imbalance and Table 3 presents the effect of matching on our estimates of treatment effect. Each row of Table 3 presents the impact estimates across alternative models for a given dependent variable. The second (baseline) column presents estimates from regressing a given dependent variable on the treatment indicator alone. The third (controls) column adds our independent variables to each regression (see Table 1 for the variable list). The final two columns (CEM-SS and CEM-KM) present our estimates when using CEM with the Shimazaki-Shinomoto and k-means coarsening, respectively. Recall that all estimates are marginal effects (evaluated at the mean) from the appropriate linear or nonlinear model. That is, OLS is used for continuous outcome variables, logit models are used for binary outcomes, Poisson or negative binomial regressions are used for count data, and Tobit models are used for censored outcomes. Also recall that all standard errors are clustered at the province level. Consider first the results for the (log of) municipal expenditures per capita variable (expenditure). While the baseline estimate shows that treated districts witnessed a statistically significant 19 percent increase in expenditures per capita, we see more subdued effects with the addition of control variables and matching. In particular, the CEM-SS and CEM-KM columns show estimated increases of 12 and 7 percent, respectively, but neither of these estimates is statistically significant. Recall, however, that FOCAM income must be directed to capital expenditures. Looking at the results for the (log of) municipal capital expenditures per capita variable (capital), we see relatively robust positive effects. Most notably, the CEM-SS and CEM-KM columns show statistically significant increases in capital of 22 and 17 percent, respectively. While treated districts thus witnessed increased capital expenditures, we conversely find evidence of a reduction in (the 14 log of) municipal current expenditures per capita (current). For example, our CEM-KM regression shows a statistically significant 11 percent reduction in current. The reduction in current expenditures rationalizes the statistically insignificant estimates for overall expenditures, but it is an unexpected result. We further analyze this result below, but first examine the estimates associated with the other dependent variables. Given the estimated positive effect on capital expenditures, it is natural to ask whether there is an associated positive effect on local infrastructure. While estimates vary across alternative models, our preferred models (CEM-SS and CEM-KM) show robust positive impacts on internet,construction,roads,sports, and cadastre. Regarding internet, for example, our CEM-SS and CEM-KM results show that treated districts were, respectively, 8 and 11 percent more likely to have access to internet in 2012. To cite another example, the CEM-SS and CEM-KM regressions also show that treated districts constructed an additional 2.09 and 2.87 square kilometers of primary roads in 2012, respectively. The other infrastructure-related results can be interpreted analogously and we leave this exercise to the reader. The final dependent variable in Table 3 to be discussed is the (log of) nighttime lights variable (lights). Given evidence of increased capital expenditure and an associated impact on local infrastructure, we might expect to see a positive impact on the nighttime lights variable. We, however, find only very limited evidence for this hypothesis. Our estimate with control variables suggests that treated districts witnessed a statistically significant 7 percent increase in lights. Our CEM-SS and CEM-KM results nevertheless show that this estimate is biased upward, and that more accurate point estimates are 2 and 4 percent, respectively. While we indeed find positive point estimates, neither the CEM-SS or CEM-KM results are statistically significant at any conventional level. Accordingly, increased capital expenditures and local infrastructure development do not appear to have translated into robust impacts on nighttime lights, though it is possible that additional impacts will emerge with the passage of more time. We conclude this section by further examining the above result that FOCAM transfers reduced municipal current expenditures per capita (current). As stated, we hypothesize that this reduction in current expenditures is related to the absorptive capacity of municipal governments. That is, we hypothesize that municipalities with low absorptive capacity coped with the increased administrative burden of FOCAM transfers by reallocating administrative effort toward executing new capital expenditures. To test our hypothesis, we thus add to previous models a term that interacts staff with our treatment indicator. Table 4 presents results from this augmented model. Full regression results are presented for the model with all 15 controls and no matching, as well as for the CEM-SS and CEM-KM approaches. Recall that all regressions are estimated with OLS using standard errors clustered at the province level. While the interaction term (treated ×staff ) is statistically insignificant in the regression without matching (second column of Table 4), the preferred CEM-SS and CEM-KM regressions show statistically significant point estimates of 0.12 and 0.13, respectively. To gain insight into the economic significance of these estimates, in Figure 4 we plot the heterogeneous treatment effects implied by the CEM-KM regression.24 That is, we plot the point estimate and 95 percent confidence interval associated with the treatment effect on current as it varies with changes in staff. The results confirm our hypothesis: districts with low levels of absorptive capacity (i.e. staff ) witness statistically significant reductions in current expenditures (i.e. current). More specifically, our results imply that districts with staff at one standard deviation below the mean witnessed an approximate 20 percent reduction in current (see Table 2 for descriptive statistics). Further, the point estimates also show that districts with high levels of staff witnessed increases in current, perhaps due to purchasing goods and services complementary to the new capital expenditures.25 6 Conclusions We examined the economic effect of windfall gains by studying the impact of the Camisea Fund for Socioeconomic Development (FOCAM) in Peru. The rules governing FOCAM allocation created a natural experiment from which we were able to identify the effect of the transfers on municipal accounts, local infrastructure, and local development. Using a newly constructed district-level dataset for the years 2005 and 2012, we first found evidence of increased capital expenditures in treated districts. Second, we found that increased capital expenditures were associated with positive impacts on local infrastructure. That is, we found evidence of positive impacts on access to internet, licenses granted for new construction, and the building of primary roads, among other things. Finally, we found that districts with low absorptive capacity coped with the increased administrative burden of FOCAM transfers by reallocating administrative effort toward (away from) executing capital (current) expenditures. This last results is particularly noteworthy, especially in the Peruvian context. The FOCAM transfer scheme is part of a larger fiscal reorganization effort in Peru that has attempted to use fiscal decentralization to reduce corruption and improve public service delivery. To this end, fiscal decentralization was accompanied 24A similar exercise could be undertaken with the CEM-SS model, but the results are virtually identical. 25Note, however, that zero always falls within the 95 percent confidence interval for the districts with high absorptive capacity. Also note that we tried a number of alternative specifications to permit some non-linearity in Figure 4. The associated hypothesis tests nevertheless revealed that the simple linear specification is most appropriate. 16 by stringent fiduciary requirements and rigid participatory budgeting guidelines for local governments and their elected officials. The anecdotal evidence suggests that many municipalities had difficulties complying with the drafted regulations and were unable to execute additional expenditures (see Loayza et al. [2014] and references therein for details). Our results, however, suggest that municipalities lacking absorptive capacity may not necessarily be unable to execute additional expenditures altogether, but may rather reallocate administrative effort to accommodate the additional expenditures. Thus, absent technical support and capacity-building efforts, such windfall gains may have unintended consequences. While we believe that our results have important policy implications for improving the efficacy of windfall gains, our analysis has some limitations that should be acknowledged. First, data availability and compatibility issues precluded us from constructing a panel dataset including interim years. Second, also due to data availability issues, we were unable to analyze the potential consequences of reduced municipal current expenditures. Finally, while we were able to identify the average treatment on the treated, we lacked the appropriate instrumental variables to estimate the elasticity of our outcome variables to FOCAM transfers (i.e. the dose-response relationship). In future research we hope to extend our analysis by remedying some of these shortcomings, particularly by identifying the dose response. 17 References Angrist, J. and Kugler, A. (2008). Rural windfall or a new resource curse? Coca, income, and civil conflict in Colombia. The Review of Economics and Statistics, 90(2):191–215. Arndt, C., Jones, S., and Tarp, F. (2010). Aid, growth, and development: Have we come full circle? Journal of Globalization and Development, 1(2). Arzaghi, M. and Henderson, J. (2005). Why countries are fiscally decentralizing. Journal of Public Economics, 89(7):1157–1189. Bazzi, S. and Clemens, M. (2013). Blunt instruments: Avoiding common pitfalls in identifying the causes of economic growth. American Economic Journal: Macroeconomics, 5(2):152–186. Becker, S., Egger, P., and von Ehrlich, M. (2013). Absorptive capacity and the growth and investment effects of regional transfers: A regression discontinuity design with heterogeneous treatment effects. American Economic Journal: Economic Policy, 5(4):29–77. Bird, R. (1993). Threading the fiscal labyrinth: Some issues in fiscal decentralization. National Tax Journal, 46(2):207–227. Bird, R. and Smart, M. (2002). Intergovernmental fiscal transfers: International lessons for developing countries. World Development, 30(6):899–912. Blackwell, M., Iacus, S., King, G., and Porro, G. (2009). cem: Coarsened exact matching in Stata. Stata Journal, 9(4):524–546. Brollo, F., Nannicini, T., Perotti, R., and Tabellini, G. (2010). The political resource curse. NBER Working Paper No. 15705. Brunnschweiler, C. and Bulte, E. (2008). The resource curse revisited and revised: A tale of paradoxes and red herrings. Journal of Environmental Economics and Management, 55(3):248–264. Burnside, C. and Dollar, D. (2000). Aid, policies, and growth. The American Economic Review, 90(4):847– 868. Caselli, F. and Michaels, G. (2009). Do oil windfalls improve living standards? Evidence from Brazil. NBER Working Paper No. 15550. Chen, X. and Nordhaus, W. (2011). Using luminosity data as a proxy for economic statistics. Proceedings of the National Academy of Sciences, 108(21):8589–8594. Contralor´ıa General de la Rep´ublica del Per´u (2014). Estudio del proceso de descentralizaci´on en el Per´u. Contralor´ıa General de la Rep´ublica, Lima, Peru. Dalgaard, C. and Olsson, O. (2008). Windfall gains, political economy and economic development. Journal of African Economies, 17(suppl 1):72–109. De Mello, L. (2000). Fiscal decentralization and intergovernmental fiscal relations: A cross-country analysis. World Development, 28(2):365–380. Dillinger, W. (1994). Decentralization and it implications for urban service delivery. Urban Management and Municipal Finance Discussion Paper No. 16, The World Bank. Dube, O. and Vargas, J. (2013). Commodity price shocks and civil conflict: Evidence from Colombia. The Review of Economic Studies, 80(4):1384–1421. Gadenne, L. and Singhal, M. (2013). Decentralization in developing economies. NBER Working Paper No. 19402. 18 Hansen, H. and Tarp, F. (2001). Aid and growth regressions. Journal of Development Economics, 64(2):547– 570. Henderson, J., Storeygard, A., and Weil, D. (2012). Measuring economic growth from outer space. The American Economic Review, 102(2):994. Holland, P. (1986). Statistics and causal inference. Journal of the American Statistical Association, 81(396):945–960. Iacus, S., King, G., and Porro, G. (2011). Multivariate matching methods that are monotonic imbalance bounding. Journal of the American Statistical Association, 106(493):345–361. Iacus, S., King, G., and Porro, G. (2012). Causal inference without balance checking: Coarsened Exact Matching. Political Analysis, 20(1):1–24. Imbens, G. and Wooldridge, J. (2009). Recent developments in the econometrics of program evaluation. Journal of Economic Literature, 47(1):5–86. Inter-American Development Bank (2002). Program for institutional strengthening and environmental and social management support for the Camisea Gas Project: Loan proposal. Retrieved from www.iadb.org/ en/projects/project-description-title,1303.html?id=PE0233. Accessed September 23, 2015. Inter-American Development Bank (2004). Informe resumen de la ejecuci´on de los compromisos ambientales y sociales del Projecto Camisea. Retrieved from idbdocs.iadb.org/wsdocs/getdocument.aspx?docnum= 455833. Accessed September 23, 2015. King, G. and Zeng, L. (2006). The dangers of extreme counterfactuals. Political Analysis, 14(2):131–159. Loayza, N., Rigolini, J., and Calvo-Gonz´alez, O. (2014). More than you can handle: Decentralization and spending ability of Peruvian municipalities. Economics & Politics, 26(1):56–78. Meyer, B. (1995). Natural and quasi-experiments in economics. Journal of business & economic statistics, 13(2):151–161. Michalopoulos, S. and Papaioannou, E. (2013). Pre-colonial ethnic institutions and contemporary african development. Econometrica, 81(1):113–152. Ministerio de Econom´ıa y Finanzas (2015). Fondo de Desarrollo Socioecon´omico de Camisea - FOCAM. Retrieved from www.mef.gob.pe/index.php?option=com content&view=article&id=451&Itemid= 100962&lang=es. Accessed September 23, 2015. Munilla, I. (2010). People, power, and pipelines: Lessons from Peru in the governance of gas production revenues. Oxfam America, Boston, MA. Paler, L. (2011). The subnational resource curse: Causes, consequences, and prescriptions. Revenue Watch Institute. Rajan, R. and Subramanian, A. (2008). Aid and growth: What does the cross-country evidence really show? The Review of Economics and Statistics, 90(4):643–665. Sachs, J. and Warner, A. (1995). Natural resource abundance and economic growth. NBER Working Paper No. 5398. Shimazaki, H. and Shinomoto, S. (2007). A method for selecting the bin size of a time histogram. Neural Computation, 19(6):1503–1527. The Economist (2003). Gas for peru v green imperialism. Retrieved from www.economist.com/node/1978816. Accessed September 23, 2015. 19 van der Ploeg, F. (2011). Natural resources: Curse or blessing? Journal of Economic Literature, 49(2):366– 420. van der Ploeg, F. and Poelhekke, S. (2010). The pungent smell of “red herrings”: Subsoil assets, rents, volatility and the resource curse. Journal of Environmental Economics and Management, 60(1):44–55. Vicente, P. (2010). Does oil corrupt? Evidence from a natural experiment in West Africa. Journal of Development Economics, 92(1):28–38. Walsh Peru S.A. (2002). EIA Gasoducto Camisea - Lima. Retrieved from www.southperupanel.org/files/ eias/03.%20CAMISEA-LIMA/VOL%20I/7.0%20Alternativas.pdf. Accessed September 23, 2015. Wang, H. and Song, M. (2011). Ckmeans.1d.dp: Optimal k-means clustering in one dimension by dynamic programming. The R Journal, 3(2):29–33. 20 -.6 -.4 -.2 0 .2 .4 .6 Estimate of treatment effect 1.5 22.5 33.5 44.5 55.5 6 Log of white-collar staff per ten thousand people Figure 4: Heterogeneity of impact on the log of municipal current expenditures per capita 27