Electoral Politics, Fiscal Policy, and the Resource Curse

Abstract

While some governments use natural resources for immediate political gain, others create transparent institutions that promote sustainable development. What explains this variation? Using novel data for Latin America between 1990 and 2019, I show that executive incumbents are more likely to restrict their discretion over natural resource revenue when public approval is high and legislative opposition is strong. When rulers are safe in their seats, they can use public funds for long-run developmental strategies, rather than short-term political survival. When there is a strong legislative opposition, rulers can signal a desire to compromise by relinquishing control over resource revenue. These findings, illustrated by the case of Mexico, suggest that a combination of high support and strong opposition provides space to create long-term fiscal policy frameworks while generating short-term incentives to do so.

Appendices

Appendix

A Countries Included in the Statistical Analysis

Argentina, Bolivia, Brazil, Chile, Colombia, Ecuador, Guatemala, Mexico, Peru, Venezuela.

B Summary Statistics

Table 4; Fig. 7.

Table 4 Descriptive statistics

Fig. 7

Correlation matrix for the main independent variables. This figure shows the correlation between the three main independent variables, indicating that they are only weakly correlated with each other

C Robustness Checks

C.1 Logistic Regressions with Cubic Polynomials

Carter and Signorino (2010) show that logistic regressions with time dummies might suffer from two problems: separation and inefficiency. Penalized maximum likelihood addresses separation concerns, as “it also produces finite parameter estimates even in the presence of quasi- or complete separation” (Cook et al. 2020, 96). Still, the results are not contingent on the use of time dummies; as Tables 5 and 6 show, models estimated with cubic polynomials are similar to the main specifications.

Table 5 Determinants of SFI creation or regulation, 1990–2019 (with cubic polynomials)

Table 6 Determinants of SFI creation or regulation: natural resource funds, 1990–2019 (with cubic polynomials)

C.2 Linear Regressions with Wild Cluster Bootstrap Standard Errors

The main analysis in Tables 2 and 3 clusters standard errors by country. However, there are only ten countries. When the number of clusters is small, clustered standard errors might lead to overly narrow confidence intervals and over-rejection (Cameron and Miller 2015). Thus, I follow Cameron and Miller (2015) and estimate models with Wild cluster bootstrap standard errors as a robustness check. The Wild bootstrap cannot be calculated for non-linear models because it requires additively separable errors, so I estimate linear regressions instead. The results of linear regressions with Wild cluster bootstrap standard errors, reported in Tables 7 and 8, are statistically and substantively similar to the main results.

Table 7 Determinants of SFI creation or regulation, 1990–2019 (with wild cluster bootstrap standard errors)

Table 8 Determinants of SFI creation or regulation: natural resource funds, 1990–2019 (with wild cluster bootstrap standard errors)

C.3 Survival Models

Natural resource policy passage is not a terminal event; countries are constantly “at risk” of experiencing this event. Ecuador, for instance, passed seven such legal documents, indicating that passing the first document does not preclude countries from passing another one. This is why Tables 2 and 3 report the results of logistic regressions, rather than survival models.

The logic of a survival model would be that once countries adopt some type of natural resource policy, they are no longer at risk of passing another such policy and exit the sample, which would not be appropriate for the context of this study. In addition, my analysis begins in 1990; given that several countries (Ecuador, Mexico, Peru, Trinidad and Tobago, and Venezuela) passed their first policy in 1999 or 2000, a survival model would lead to a considerable loss of information. Indeed, a Cox proportional hazards model with all key independent variables and control variables does not converge because there are not enough observations.

As an imperfect solution, Table 9 presents the results of bivariate Cox proportional hazards models, combining each key independent variable (Executive approval, Opposition vote share, or Number of protests) to each outcome of interest (the time until the first document is observed). These models are by no means ideal, but provide suggestive evidence that passing the first natural resource policy and passing any natural resource policy are decisions that might be driven by similar factors.

Table 9 Determinants of SFI creation or regulation, 1990–2019 (Cox proportional hazards models)

C.4 Models Excluding Outliers for Number of Protests

Tables 10 and 11 re-estimate some of the main models, excluding outliers for Number of protests (that is, country-quarters that experienced over six protests). Tables 10 and 11 provide some reassurance that this is not the case, as the results are robust to dropping these observations.

Table 10 Determinants of SFI creation or regulation, 1990–2019 (excluding extreme values for protests)

Table 11 Determinants of SFI creation or regulation: natural resource funds, 1990–2019 (excluding extreme values for protests)

C.5 Models Interacting Public Support With Political Opposition

Table 12 presents the results of models that interact public support (measured as Executive approval) with political opposition (measured either as Opposition vote share or as Number of protests). Figures 8 and 9 plot the marginal effects of these interaction terms on Any document, suggesting that high executive support and low executive discretion jointly are significantly associated with an increase in the odds of passing any SFI-related document. Figures 10 and 11 plot the marginal effects of these interaction terms on Fund document; in this case, the interactive effects are far weaker.

Table 12 Determinants of SFI creation or regulation, 1990–2019 (with interactions between public support and political competition)

Fig. 8

Predicted probability of Any document at differentvalues of public approval, conditional on Opposition Vote Share. Based on Model 1 of Table 12(re-estimated without country fixed effects), these figures simulate the predicted probability of observing Fund document, with 95% confidence intervals, at different values of Executive approval, conditional on Opposition vote share at its minimum (a), median (b), mean (c), or maximum (d) value. The remaining variables are held at their means (with dichotomous variables held at zero)

Fig. 9

Predicted probability of Any document at different values of public approval, conditional on Number of Protests. Based on Model 2 of Table 12 (re-estimated without country fixed effects), these figures simulate the predicted probability of observing Fund document, with 95% confidence intervals, at different values of Executive approval, conditional on Number of protests at its four most frequent values (0, 1, 2, and 3). The remaining variables are held at their means (with dichotomous variables held at zero)

Fig. 10

Predicted probability of Fund document at different values of public approval, conditional on Opposition Vote Share. Based on Model 3 of Table 12 (re-estimated without country fixed effects), these figures simulate the predicted probability of observing Fund document, with 95% confidence intervals, at different values of Executive approval, conditional on Opposition vote share at its minimum (a), median (b), mean (c), or maximum (d) value. The remaining variables are held at their means (with dichotomous variables held at zero)

Fig. 11

Predicted probability of Fund document at different values of public approval, conditional on Number of Protests. Based on Model 4 of Table 12 (re-estimated without country fixed effects), these figures simulate the predicted probability of observing Fund document, with 95% confidence intervals, at different values of Executive approval, conditional on Number of protests at its four most frequent values (0, 1, 2, and 3). The remaining variables are held at their means (with dichotomous variables held at zero)

C.6 Models with Lagged Executive Approval

Tables 13 and 14 examine the effect of the independent variable Executive approval when lagged at one to five quarters. The results are robust to these changes. In fact, Executive approval has the largest effect on Any document at time $t-2$ and on Fund document at time $t-4$; both effects are statistically significant.

Table 13 Determinants of SFI creation or regulation, 1990–2019 (with lagged executive approval)

Table 14 Determinants of SFI creation or regulation: natural resource funds, 1990–2019 (with lagged executive approval)

C.7 Models With Executive Opposition Vote Share

Lastly, Table 15 tests for the effect of Executive opposition vote share, measured as the vote share of all opposition candidates in the first (or only) round of presidential elections. This variable has a small and inconsistent effect on SFI creation and regulation, suggesting that the mechanism at play is not only the existence of an opposition, but a standing opposition.

Table 15 Determinants of SFI creation or regulation, 1990–2019 (with executive opposition vote share)