Chapter

IV Quantitative Assessment of the Impact of Fiscal Institutions on Fiscal Outcomes

Author(s):
Mauricio Villafuerte, Rolando Ossowski, Theo Thomas, and Paulo Medas
Published Date:
April 2008
Share
  • ShareShare
Show Summary Details

This section presents a quantitative approach to assess the impact of SFIs on the broad fiscal policy responses of oil-producing countries to the oil boom up to 2005. As discussed in Section II, the responses of countries to the oil boom have varied significantly. This section assesses the extent to which SFIs (oil funds and fiscal rules) can explain differences in fiscal responses across countries and over time. In addition, insofar as possible, the impact of broader institutional frameworks on fiscal outcomes is also assessed.

The econometric analysis focuses on the quantitative impact of SFIs on key fiscal variables, controlling for differences in the size of government net wealth and the degree of dependence on oil revenue. Many countries have introduced SFIs in an attempt to influence the level and dynamics of expenditures, and in particular to limit the response to fluctuations in oil revenue in any given year.1 As such, it would be expected that on average countries with SFIs would have a more restrained fiscal response to the oil boom, particularly as it was not known how much of the oil price increase would be long-lasting. Nevertheless, when measuring the impact of SFIs it is necessary to take country-specific factors into account. As such, the econometric analysis is structured in the following manner:

  • Regressions were run to assess the impact of SFIs on key fiscal variables; namely, the change in the non-oil primary balance ratio to non-oil GDP, the average growth of government expenditures in real terms, and the correlation between expenditure and oil revenue.
  • The analysis focuses on assessing whether the presence of SFIs changes a country’s fiscal response. The regressions control for key economic variables in oil-producing countries, in particular the size of government net wealth and the degree of dependence on oil revenue. The objective is to assess whether different fiscal outcomes are observed in a country with SFIs compared to a similar country without SFIs. Other control variables that may influence the fiscal response (such as per capita GDP) are also included.2
  • The quantitative assessment also looks at the more direct effect of SFIs in affecting the response to changes in oil revenue and the size of government net wealth.

Non-Oil Primary Balance

The empirical results suggest that SFIs have not had a discernible impact on the fiscal position, as measured by the ratio of the non-oil primary balance to non-oil GDP, which tends to be more dependent on the short-term volatility of oil revenue. As Table 4.1 shows, the results indicate that the introduction of SFIs would not have affected the non-oil primary deficit, even after controlling for government net wealth and the degree of dependence on oil revenue (the ratio of oil revenue to total revenue).3 The impact of SFIs is negative under all alternative specifications during the oil boom period, suggesting that the presence of SFIs is associated with a lower non-oil primary balance; however, the coefficients do not tend to be statistically significant.4 These findings are also confirmed for the longer sample (Table 4.1, columns 4–7).5 In general, however, the estimates are not statistically significant.6 The evidence also suggests that liquidity considerations are a key factor for fiscal policy—that is, the size of the non-oil primary balance depends mostly on the (annual) share of oil revenue in total revenue and changes in oil prices (Table 4.1). In general, government net wealth does not appear to be influencing fiscal policy.

Table 4.1.Dependent Variable: Non-Oil Primary Balance
Sample: 2000–05Sample: 1992–2005
OLSFixed effectsFixed effectsArellano-Bond
(1)(2)(3)(4)(5)(6)(7)
Log (GDP per capita in US$, lagged)−0.2423.8**22.8**−3.27
(−0.19)(2.11)(2.13)(−0.91)
Net government wealth−0.001−0.010.010.002***0.002***−0.0010.0
(as a share of non-oil GDP)(−1.18)(−1.26)(0.84)(2.92)(3.00)(−0.74)(0.47)
SFIs dummy−4.69−2.26−0.8−0.5−0.2−1.5
(−1.00)(−0.84)(−0.73)(−0.45)(−0.14)(−0.99)
Oil revenue (as a share of total revenue)−1.1***−0.56***−0.59***−0.30***−0.31***−0.41***−0.33***
(−7.50)(−3.19)(−3.12)(−5.46)(−5.62)(−5.71)(−5.31)
Oil price (lagged)0.1−0.68**−0.68**0.00.04−0.16−0 .17**
(0.30)(−2.28)(−2.28)(−0.55)(0.50)(−1.35)(−2.12)
Inflation (lagged)−0.12**−0.01−0.01
(−2.49)(−0.13)(−0.15)
SFI* net wealth−0.015
(−1.17)
Non-oil primary balance (as a share of non-oil GDP, lagged)0.60***0.28***
(3.43)(6.48)
Composite index of institutional0.3***
quality (ICRG)(2.68)
Democratic accountability−1.65***−1.5**
(−3.33)(−2.21)
Bureaucratic quality1.452.06*
(1.15)(1.73)
Government stability0.60**
(2.34)
Law and order2.2***1.77*
(2.76)(1.84)
Number of observations172166166284284268255
Source: IMF staff calculations.

Note: T-statistic is in parentheses, with *** = significant at 1 percent; ** at 5 percent; * at 10 percent. The regressions use robust standard errors. The constant term is not shown. Column 1 is based on standard ordinary least squares (OLS), whereas columns 2–5 correct for fixed effects. Based on the Hausman specification test, fixed effects are preferred to random effects regressions (not shown). Columns 6 and 7 used a dynamic specification developed by Arellano and Bond (1991) to correct for possible endogeneity of the independent variables. ICRG is the International Country Risk Guide.

Source: IMF staff calculations.

Note: T-statistic is in parentheses, with *** = significant at 1 percent; ** at 5 percent; * at 10 percent. The regressions use robust standard errors. The constant term is not shown. Column 1 is based on standard ordinary least squares (OLS), whereas columns 2–5 correct for fixed effects. Based on the Hausman specification test, fixed effects are preferred to random effects regressions (not shown). Columns 6 and 7 used a dynamic specification developed by Arellano and Bond (1991) to correct for possible endogeneity of the independent variables. ICRG is the International Country Risk Guide.

In contrast to SFIs, broader governance institutions do seem to have an impact on the non-oil primary balance (Table 4.1, columns 4, 5, and 7). Once broader indicators of institutional quality are introduced, the SFIs’ coefficients tend to zero and become highly insignificant, while higher institutional quality seems to be associated with lower non-oil deficits. A composite indicator of the overall quality of institutions (column 4) suggests that the higher the quality of institutions, the higher the non-oil primary balances.7 Concerning individual indices included in the composite indicator (column 5), the main contributors for the higher non-oil primary balance are government stability, law and order, and bureaucratic quality. Democratic accountability has a negative impact on the non-oil primary balance.8 The results are robust to controlling for possible dynamic effects (columns 6 and 7).

Public Spending

The evidence indicates that SFIs did not have a significant impact on expenditure growth nor helped constrain the expenditure response to changes in oil revenue. The econometric results (Table 4.2), using various specifications, show that the impact of SFIs on expenditure tends to have the “wrong” sign (positive), but the coefficients are not statistically significant after controlling for fixed effects (columns 2 and 5).9 When testing for the interaction between the SFI variable and oil revenue growth (columns 3 and 6)—to better assess the ability of SFIs to affect the expenditure sensitivity to contemporaneous changes in oil revenue—the results are mixed.10 During the boom period, SFIs appear to strengthen the response of spending to rising oil revenue. In the longer sample, this relationship is not statistically significant. These results suggest that SFIs have not moderated spending responses to oil revenue volatility, particularly in periods of large changes. In addition, the data show that government net wealth is not significant because expenditures tended to react more strongly to changes in oil revenue.11

Table 4.2.Dependent Variable: Expenditures

(Annual real growth rate)

Sample: 2000–05Sample: 1992–2005
OLSFixed effectsOLSFixed effects
(1)(2)(3)(4)(5)(6)
Log of expenditures (as a share of GDP, lagged)−13.7**−49.3***−53.6***−17.8***−35.2***−35.4***
(−2.30)(−3.59)(−3.97)(−4.65)(−5.37)(−5.50)
Log (GDP per capita in US$, lagged)−1.63*−4.55−3.04−0.58−8.8**−7.7**
(−1.78)(−0.70)(−0.49)(−0.71)(−2.2)(−1.99)
Net government wealth (as a share of non-oil GDP)0.0001−0.003***−0.004***0.000.000.00
(0.45)(−2.75)(−3.49)(−0.53)(−0.05)(−0.2)
Oil revenue growth (percent change)0.06***0.07**0.050.08***0.08***0.073***
(2.83)(2.24)(2.04)(3.55)(3.20)(2.68)
SFIs dummy5.58**8.014.32**3.13
(2.40)(1.13)(2.36)(1.08)
Inflation (lagged)−0.0160.040.08−0.23***−0.28***−0.29***
(−0.30)(0.64)(1.06)(−2.63)(−2.67)(−2.86)
Oil revenue (as a share of total revenue, lagged)0.2***0.6***0.71***0.18***0.43***0.45***
SFIs dummy* oil revenue growth(3.2)(2.95)(3.44)(3.81)(3.58)(3.80)
0.07**0.02
(2.25)(0.62)
Number of observations171171171297297297
Source: IMF staff calculations.

Note: T-statistic is in parentheses, with *** = significant at 1 percent; ** at 5 percent; * at 10 percent. The regressions use robust standard errors. The constant term is not shown. Based on the Hausman specification test, fixed effects are used instead of random effects regressions (not shown).

Source: IMF staff calculations.

Note: T-statistic is in parentheses, with *** = significant at 1 percent; ** at 5 percent; * at 10 percent. The regressions use robust standard errors. The constant term is not shown. Based on the Hausman specification test, fixed effects are used instead of random effects regressions (not shown).

The empirical analysis also fails to show that SFIs have had an impact on the correlation between expenditure and oil revenue during the oil boom (Table 4.3). The latter is proxied by the ratio between changes in expenditures and changes in oil revenue, where the strongest impact from fiscal institutions would be expected to be found.12 When the corruption index is introduced, the coefficient of the SFI variable becomes close to zero and highly insignificant. The coefficient for the (lack of) corruption index, however, is significant and suggests that countries with lower levels of corruption have lower correlations between expenditures and oil revenue.13 These results seem to be in line with the “voracity effect” discussed in the literature, where countries with the weakest institutions tend to spend more during revenue windfalls (Tornell and Lane, 1999), and the evidence showing that many countries with the lowest government effectiveness indices (as measured by Kaufmann, Kraay, and Mastruzzi, 2005) are among those spending the most.

Table 4.3.Dependent Variable: Ratio of the Change in Expenditure to the Change in Oil Revenue
Sample: 2000–05
OLSFixed effects
(1)(2)(3)
Log (GDP per capita, lagged)−0.213.04.6
(−0.73)(0.87)(1.17)
Net wealth as a share of non-oil GDP (lagged)0.000.00.0
(−1.14)(−1.05)(−0.38)
Oil revenue as a share of total revenue−0.034−0.2−.25*
(−1.00)(−1.42)(−1.75)
SFIs dummy–1.11–4.4−3.7
(−0.67)(−0.99)(−0.84)
Corruption index−3.96**
(−2.08)
Democratic accountability−2.26*
(−1.77)
Number of observations170170156
Source: IMF staff calculations.Note: T-statistic is in parentheses, with *** = significant at 1 percent; ** at 5 percent; * at 10 percent. The regressions use robust standard errors. The constant term is not shown. Based on the Hausman specification test, fixed effects are used instead of random effects regressions (not shown).
Source: IMF staff calculations.Note: T-statistic is in parentheses, with *** = significant at 1 percent; ** at 5 percent; * at 10 percent. The regressions use robust standard errors. The constant term is not shown. Based on the Hausman specification test, fixed effects are used instead of random effects regressions (not shown).
1This tends to reflect short-term considerations (such as preventing volatile spending and enhancing macro-fiscal management), and/or longer-term objectives (such as promoting sustainability and intergenerational equity).
2Appendix IV describes in more detail the econometric methodology and data used. In particular, the appendix describes the strategy used to address some potential key econometric problems, mainly the following: (i) countries with relatively large non-oil deficits or difficulties in containing spending may be the ones more likely to introduce SFIs, which could lead to biased estimates, because the SFIs would “appear” to cause higher deficits or expenditure growth; (ii) SFIs could be influenced by the dependent (fiscal) variables; and (iii) it may be difficult to distinguish the impact of the introduction of the SFIs in countries where their introduction overlapped with the beginning of the boom.
3The SFI variable takes a value of 1 if the country has an oil fund or a fiscal rule/FRL, and zero otherwise. The analysis does not look separately at different SFIs owing to data limitations. In addition, introducing variables that include a more qualitative assessment of SFIs (for example, differentiating SFIs by quality of design or actual specific objectives) could bias the estimates. The preceding section presents a more qualitative discussion of the SFIs.
4To test whether the presence of SFIs would constrain the impact of changes in government net wealth, Table 4.1, column 3 included the interaction between the SFI variable and government net wealth, which is also not significant.
5The main results are robust to controls for possible endogene-ity of the SFI variable. The coefficients of the SFI are small and still not statistically significant at the 10 percent level (Table 4.1, columns 6–7).
6The regressions, except for column 1, exclude Equatorial Guinea, because the very large increase in oil production and revenue in the late 1990s would tend to dominate the results. If included, the regressions would tend to show a more negative impact of fiscal institutions.
7The indicator of institutional quality is proxied by the International Country Risk Guide’s political risk index. GDP per capita is omitted from the regressions that include some of the institutional variables to avoid multicollinearity.
8As discussed in the literature, the channels through which the different indicators of institutional quality affect fiscal policy can be complex and difficult to capture in regressions. In particular, some studies have suggested that countries that move toward more democratic regimes may spend more in the first years of the transition; the regression may be capturing this phenomenon. See Fabrizio and Mody (2006) and Manasse (2006).
9The coefficient of the SFIs is positive and significant when using standard ordinary least squares (OLS) regressions.
10The interaction between SFIs and government net wealth was also tested, but was not statistically significant.
11Columns 4–6 exclude Angola, because the period of high inflation distorted the estimation of real variables and the econometric results.
12Coefficients for the SFI dummy tend to have the expected sign—the existence or introduction of SFIs reduces the correlation between expenditures and oil revenue. The estimates, however, are not significant.
13The presence of democratic accountability also appears to help contain the correlation between expenditures and growth. The impact of the institutional variables is robust to excluding potential outliers (e.g., Norway).

    Other Resources Citing This Publication