Helping Countries Develop
Chapter

3 Persistence of Fiscal Adjustments and Expenditure Composition in Low-Income Countries

Author(s):
Benedict Clements, Sanjeev Gupta, and Gabriela Inchauste
Published Date:
September 2004
Share
  • ShareShare
Show Summary Details
Author(s)
Emanuele Baldacci, Benedict Clements, Sanjeev Gupta and Carlos Mulas-Granados 

There is a consensus among researchers that fiscal consolidations need to persist if they are to have a positive effect on growth. In general, the persistence of high-quality fiscal adjustment can improve macroeconomic stability and reduce expectations that higher taxes and interest rates will be needed in the future to finance fiscal imbalances. Short-lived fiscal consolidations, on the other hand, can be harmful for growth, as they signal that the initial improvement in the fiscal budget cannot be maintained and could be reversed in the medium term. An understanding of what makes fiscal consolidations sustainable is therefore essential to unraveling how fiscal adjustment influences growth (von Hagen and Strauch, 2001).

This chapter assesses what are the factors determining the end of a fiscal consolidation episode and analyzes whether improvements in the composition of public expenditure help fiscal adjustments in low-income countries last longer. Many studies have analyzed episodes of fiscal adjustment in industrial countries (for example, Alesina and Perotti, 1995), but similar studies for developing countries—and in particular low-income countries—are lacking.

The methodology traditionally used to assess the sustainability of fiscal adjustment episodes over time focused on large fiscal consolidation spells. Data on fiscal consolidation periods were used to analyze the features of successful fiscal adjustments, defined as those cases where fiscal control was maintained over an adequate period of time (Alesina and Ardagna, 1998). A number of recent papers (Maroto and Mulas-Granados, 2001; Gupta and others, 2002; von Hagen, Hughes Hallett, and Strauch, 2002; and Adam and Bevan, 2003) have chosen to treat the persistence of fiscal consolidations over time as endogenous, in contrast to previous studies. In this framework, survival analysis is the appropriate statistical method to assess which factors affect the persistence of fiscal consolidations. This method is superior, as it allows for a multivariate analysis of the determinants of the persistence of fiscal adjustments and makes use of all the information available in the data, rather than constraining the analysis to consolidation episodes only.

We use survival analysis to assess the factors determining the duration of fiscal consolidations in a sample of low-income countries. We expand the analysis of Gupta and others (2002) by using both a semi-parametric and a parametric approach (following Adam and Bevan, 2003) to modeling the risk of ending a fiscal adjustment spell. The results of this study confirm the previous findings for industrial countries. In particular, there is a strong link between public expenditure reform and fiscal adjustment sustainability. Fiscal adjustments that protect capital outlays are more sustainable—that is, less likely to be aborted. The persistence of fiscal adjustment is also influenced by a country’s initial fiscal conditions and by the size of the consolidation effort.

Fiscal Adjustment and Survival Analysis

The composition of fiscal adjustment is critical for the persistence of consolidation episodes. Many studies have analyzed episodes of fiscal adjustment in industrial countries, including Alesina and Perotti (1995), Alesina and Ardagna (1998), and Alesina, Perotti, and Tavares (1998). These studies focused on episodes of successful and unsuccessful fiscal consolidations. The main conclusion of these studies is that fiscal adjustments that rely primarily on reducing outlays on transfers and the wage bill are more likely to be sustainable than those based on tax increases and cuts in capital spending. Ardagna (2001) replicates these empirical results using a dynamic general equilibrium model calibrated with averaged data from 10 European economies for 1965-95. Her results indicate that fiscal stabilizations that rationalize public employment can stimulate the economy, provided that public employment does not have a positive effect on the productivity of capital and labor.

This chapter follows a different approach, consistent with recent research on this subject. Following von Hagen and Strauch (2001) and others, we define fiscal adjustments as sustainable if they persist over an adequate period of time. This definition of sustainability is somewhat different from the more common use of the term. In general, the term “sustainability” refers to whether the current fiscal stance is consistent with a non-increasing ratio of public debt to GDP over time (see Ize, 1991).

Most empirical studies on the sustainability of fiscal consolidations have used a descriptive and indirect approach to measure the determinants of sustainable fiscal adjustments. The approach consists of a two-step procedure: first, the authors preselect consolidation episodes according to a predefined threshold; and second, they provide a description of their main characteristics. As noted earlier, survival analysis is the superior technique. As such, this technique can be seen as a generalization of the previous approaches, which are based on fiscal adjustment episodes.

A few studies have applied this analysis to a wide set of countries. Maroto and Mulas-Granados (2001) and von Hagen, Hughes Hallet, and Strauch (2002) defined fiscal adjustments as sustainable if they persist over an adequate period of time, and applied this methodology to a sample of industrial countries. Gupta and others (2002; 2004) used a duration analysis to assess the factors underlying sustainable fiscal adjustment in 39 low-income countries. Adam and Bevan (2003) expanded the duration analysis of fiscal adjustments to a sample of 127 countries, including industrial, emerging, and developing countries (with the exception of transition economies) for the period 1970-2001.

One paper (Adam and Bevan, 2003) also provides a useful typology of the definition of “fiscal consolidation” in applying survival analysis: (1) the “level” approach (a specified threshold for the deficit); (2) the “gradient” approach (reduction of the deficit at some specified minimum rate); and (3) the composite approach. Each approach has its own strengths and weaknesses. While Adam and Bevan prefer the level approach, we define consolidations based on the gradient approach, but we condition the estimates on the existing level of deficit in each adjustment period, thereby capturing in part the “level” effect.

Data

Data for the variables used in the empirical application were drawn from the World Economic Outlook database, as well as a database for 39 countries supported by the Enhanced Structural Adjustment Facility (ESAF) or the Poverty Reduction Growth Facilty (PRGF) during the period 1990–2000.1 The fiscal policy stance is measured by the general government budget balance on a cash basis. This is defined as total revenues and grants minus total expenditures and net lending.2 A positive change in the budget balance can be interpreted as a consolidation, and a negative change as an expansion. Deficits were generally reduced during the period, with an average annual improvement of approximately ½ percentage point of GDP.

Fiscal deficits are also used to identify “post-stabilization” countries. Post-stabilization countries are defined as those that had an average budget deficit (after grants) below 2.5 percent of GDP in the 1990–2000 period.3 Based on this criterion, only seven countries can be considered post-stabilizers (Benin, The Gambia, Lesotho, Macedonia, FYR, Mauritania, Senegal, and Tanzania).

Fiscal adjustment periods are based on the observed change in the fiscal deficit as a share of GDP. Based on annual budget balance data, we generate a dummy variable called failure, which takes a value of zero when the annual variation of the budget balance is above 1½ percentage points of GDP (years of fiscal consolidation), and takes a value of one when the annual change is equal or lower than this threshold (lack of adjustment). Note that this criterion is arbitrary. One could define as a fiscal consolidation any year when a positive change in the budget balance is observed. One reason to use the threshold mentioned above, however, is to avoid labeling as “fiscal consolidations” years in which minor improvements of the budget balance took place, reflecting unintended variations of the budget, or measurement errors.4

This definition makes our results broadly comparable with previous empirical studies. For example, Alesina and Perotti (1995); Perotti (1998); and von Hagen and Strauch (2001) define episodes of fiscal consolidation as those periods in which the fiscal impulse (measured by the average cyclically adjusted primary balance) falls by at least 1¼ percent of GDP over two consecutive years, or when it increases by more than 1½ percent of GDP in one year. A successful adjustment is defined by two alternative conditions: (1) the fiscal impulse in the three years after the consolidation remains on average 2 percent of GDP above the level achieved in the last year of consolidation; or (2) the ratio of public debt to GDP three years after the consolidation is at least 5 percent of GDP below the level observed in the last year of consolidation.

Using the dates in which a failure event occurs, we create a new variable called duration, which counts the intervening years between two consecutive failures—that is, the time span that the fiscal consolidation lasts. Under the definition of consolidation described above, the minimum length of an adjustment is one year, while the maximum length is five years. The average probability of ending a consolidation is 47 percent and the average duration of a fiscal adjustment is slightly above one year.

Survival Analysis Methodology

The duration data used in this study can be summarized using three variables: the hazard rate, the survival rate, and the cumulative failure rate. The unconditional hazard function expresses the relative risk that a fiscal consolidation ends at time t, provided it was still ongoing in the previous period. The hazard function (Kaplan and Meier, 1958) is calculated as follows:

where dt represents the number of failures registered in moment t, and nt is the surviving population in moment t, before the change in status (e.g., the end of the consolidation) takes place. Intuitively, this is the failure ratio. From the hazard function, it is possible to obtain the cumulative hazard function with an estimation procedure proposed by Aalen (1978). This hazard function is given by the following expression:

The Kaplan-Meier survivor function for duration t is calculated as the product of one minus the existing risk until period t:

In the context of a non-parametric analysis, Equation (1) is sufficient to construct a life table where the initial sample of individuals at risk of experiencing the event under study is subject to the duration-specific failure rate. From the life table, summary information can be obtained of the survival process, such as the number of fiscal adjustment episodes that persist after a given number of years, the associated failure probability, and the cumulative survivor ratio up to that point. The equality of two or more survival functions across groups of countries can be formally tested using an extension of the Mantel-Haenszel test or the generalized Wilcoxon test (Cleves, Gould, and Gutierrez, 2002).5

Although non-parametric analysis is usually informative of the duration process, it cannot help assess the factors underlying the persistence of fiscal adjustments. To analyze these factors one has to link a set of covariates to the hazard function. In the literature, two different classes of models have been used: semi-parametric and parametric. Semi-parametric models do not require specific assumptions on the form of the underlying baseline hazard ratio, while parametric models usually require the advance knowledge of the shape of the hazard function.

A semi-parametric model that has been widely used in empirical studies to estimate the effects of covariates on the hazard function is the Model of Proportional Hazard (PH), which assumes that the hazard function can be described as follows:

where h0(t) is the baseline hazard function and g(X) is a function of individual covariates. This is usually defined as g(X) = exp(X′(β). Note that in this proportional specification, regressors rescale the conditional probability of ending the period of fiscal consolidation. This model can be estimated without imposing any specific functional form to the baseline hazard function, following Cox (1972).6

A fundamental property of Cox’s model is that when comparing two individuals with a different set of covariates the ratio of their hazard function is independent from the baseline hazard rate, which does not need to be specified. Model (6) can be estimated by a maximum likelihood estimator. Once the parameter is estimated, the model’s proportional hazard assumption can be tested using a generalization of the Grambsch and Therneau test involving the calculation of Schoenfeld residuals (Cleves, Gould, and Gutierrez, 2002).7

An alternative specification can be obtained by imposing one specific parametric form to the baseline hazard function h0(t). When there is sufficient deductive information on the shape of the hazard function, a parametric specification yields more efficient estimates than the Cox’s proportional hazard model. If, however, the information of the hazard function is such that the misspecification error could be large, results from a semi-parametric model would be more robust.

In the case of parametric models, the functional form most commonly assumed for the hazard function is the Weibul distribution. The baseline hazard function under this model is

where p is a parameter to be estimated and (β0 is a constant parameter. The hazard functions can be written as follows, assuming a proportional hazard specification:

When p = 1, this specification is equal to the exponential distribution that assumes the absence of any dependency on duration. The conditional probability of failure in a given interval is the same, regardless of when the observation is made. When p > 1, there is a positive duration dependency, and a negative one when p < 1. Therefore, by estimating p, it is possible to test the hypothesis of duration dependency during fiscal consolidations. An additional parametric function that has been widely used in biomedical studies is the Gompertz distribution. This distribution is suitable for modeling data with monotone hazard rates. Under this distribution, the baseline hazard function is h0(t) = exp(γt)exp(β0) and the baseline hazard function is increasing if γ is positive. Estimates of the parameters can be obtained by maximizing the corresponding log-likelihood function. Comparisons among nested models can be carried out using likelihood ratio test statistics.8 An additional advantage of this class of duration models is that frailty models describing heterogeneity can be built in this formulation (Hosmer and Lemeshow, 1999).

Empirical Results

Non-Parametric Analysis

In Table 1 we report the survival function, the hazard function, and the cumulative failure function for our sample, together with the corresponding standard errors and confidence intervals. According to the results, only 43 percent of the fiscal adjustment periods last up to the end of first year. The confidence interval puts this risk in the range between 37 percent and 48 percent owing to the presence of large dispersion in the episodes. The relative risk that a consolidation episode is discontinued at the end of the first period (the hazard rate) is only 10 percent, but it increases rapidly to 71 percent at the beginning of the second period. Finally, in the third period more than 80 percent of the adjustment episodes have already been reversed. The cumulative hazard function increases rapidly from the first to the second period, it stabilizes between the second and the third period, and it increases steadily in the remaining intervals. These results can be visually summarized in Figure 1, which shows a large share of the consolidation episodes lasting about a year, with the hazard rate peaking at the end of the first year and then rapidly declining toward zero.

Semi-Parametric Analysis

This subsection reports the results of the semi-parametric analysis on the determinants of fiscal adjustment persistence. We use Cox’s proportional hazard model described in the previous section to estimate the factors affecting the duration of fiscal adjustment efforts. In doing so we need not formulate any assumption on the functional form of the baseline hazard rate. This is allowed to vary freely with time. The actual hazard rate is thus dependent on the vector of covariates and on the estimated parameters. These can be interpreted similarly to a standard regression analysis as the effects of a unit change in the regressor on the logarithm of the probability of ending a fiscal consolidation spell. An alternative way to interpret the estimated coefficients is in terms of hazard ratios. For a given categorical variable, the hazard ratio represents the relative risk of experiencing the failure event.9

Table 1.Sustainability of Fiscal Consolidations in Low-Income Countries: Descriptive Results
Survival Function
IntervalEstimateStandard error95% confidence interval
010.9030.0150.8690.928
120.4280.0270.3740.481
230.1970.0260.1480.250
340.0630.0210.0300.112
450.000
Source: Authors’ calculations.
Cumulative Failure
IntervalEstimateStandard error95% confidence interval
010.0970.0150.0720.131
120.5720.0270.5190.626
230.8040.0260.7500.852
340.9370.0210.8880.970
451.000
Source: Authors’ calculations.
Hazard Function
IntervalEstimateStandard error95% confidence interval
010.1020.0160.0700.134
120.7130.0540.6080.818
230.7420.1020.5430.941
341.0320.2210.5991.466
452.0000.0002.0002.000
Source: Authors’ calculations.
Source: Authors’ calculations.

We regress the probability of interrupting a fiscal adjustment on a set of variables that, according to the literature, are likely to have an effect on the duration of the adjustment. The fiscal variables are as follows.

  • The size of the adjustment, measured as the cumulative change in the budget balance during the entire period of analysis. The larger the size of the consolidation, the longer the effort is hypothesized to last. In fact, a larger adjustment size signals a willingness to bring fiscal policy onto a sustainable path.

  • The composition of government spending, including both the share of current spending in total government spending and the share of transfers in current spending. The composition of the adjustment is assumed to have a critical role in the persistence of the consolidation. Fiscal adjustments based on curtailing current expenditure have been found to be more sustainable than those based on reduced capital outlays in the empirical literature on industrial countries.

  • The initial level of the fiscal deficit and the change in tax revenues and social spending, all expressed as ratios to GDP. These variables control for initial fiscal conditions and the contributions of investment in human capital and accelerated improvements in tax collection to the consolidation effort. In particular, the social spending variable is a proxy for how willing the government is to support pro-poor spending and garner broad support for the adjustment process. As such, these variables account for the possible trade-off between fiscal consolidation and the need to protect the poor from the possibly negative effect of government spending cuts.

  • We also include in the regression the change in per capita GDP growth and the previous number of failures in the adjustment process in the period considered; this is meant to control for the effect of exogenous growth shocks and past adjustment performance at the country level.

We use this model to estimate three alternative specifications: (1) we include the effect of the change in external financing as a share of GDP to take into account the effect of mostly concessional borrowing on the probability of ending an adjustment period (Model 1); (2) we omit any variable related to the composition of financing (Model 2); and (3) we include the change in the ratio of domestic financing to GDP (Model 3).

Results are reported in Table 2. The overall fit of the three models is good, although the goodness of fit indicator ranges between 21 percent for Model 1 to 16 percent for Model 3 owing to the large share of censored observations. The specification test rejects the hypothesis of omitted regressors at the 10 percent critical level and the generalized Grambsch and Therneau test based on Schoenfeld residuals confirms that the proportional hazard assumption is valid for the models presented here.

Figure 1.Sustainability of Fiscal Conditions in Low-Income Countries: Hazard and Survival Functions

Source: Authors’ calculations.

Table 2.Sustainability of Fiscal Consolidations and Budget Composition in Low-Income Countries: Results from Cox Proportional-Hazard Model, 1990-20001
Model 1Model 2Model 3
Coefficientz-testCoefficientz-testCoefficientz-test
Size of adjustment−0.04−3.06***−0.03−3.67*−0.043.68***
Initial deficit0.011.200.011.600.021.74*
Change in growth−0.02−1.99**−0.02−2.42**−0.02−1.95*
Change in social spending/GDP−0.04−1.070.010.080.020.47
Number of previous failures0.013.75***0.014.46***0.014.01***
Change in tax revenues/ GDP−0.08−2.51**−0.11−4.71***−0.061.82*
Change in transfers/ current spending0.032.18**0.021.72*0.021.79*
Change in current/ total spending0.123.92***0.125.05***0.113.13***
Change in external financing/GDP0.074.22***
Change in domestic financing/GDP0.010.07
Number of episodes167188167
Number of failures107118107
Time at risk239272239
Log likelihood−467.43−532.03−472.07
Wald test86.6275.6665.24
Probability0.000.000.00
Pseudo-R20.210.170.16
Omitted variables test−0.03−0.21−0.21−1.01−0.22−0.95
Grambsh-Therneau test χ26.291.291.36
Note: Significance at 10, 5, and 1 percent levels is indicated by *, **, and ***, respectively.

Maximum likelihood estimates with robust standard errors.

Note: Significance at 10, 5, and 1 percent levels is indicated by *, **, and ***, respectively.

Maximum likelihood estimates with robust standard errors.

The reallocation of current expenditures to capital outlays is positively related to the persistence of the adjustment in all models. Large levels of wages and salaries, transfers, and subsidies increase the probability of ending a fiscal adjustment. At the same time, allocating more public spending on capital outlays is not harmful for the sustainability of adjustment. This may be due to the positive effects of these expenditure reallocations on growth (Chu and others, 1995). Reallocating current spending away from transfers and subsidies has a positive impact on the probability of continuing the fiscal consolidation effort, while spending more on health and education is not harmful to the persistence of the adjustment.

The size of the fiscal adjustment effort also matters. The coefficient for the size of the adjustment is negative and highly significant. Thus, there appears to be little evidence of “adjustment fatigue”: countries with larger cumulative reductions in the deficit are less likely to abandon their adjustment efforts than others. This may reflect the fact that larger fiscal adjustments—including those secured in the past—signal the commitment of the authorities to continue the fiscal consolidation process.

Initial fiscal conditions are also important for the persistence of fiscal consolidations. A country with unfavorable initial fiscal conditions is more likely to end a fiscal consolidation; furthermore, a history of past failures at fiscal consolidation also foreshadows failure. This result is consistent with the findings for a sample of low-income countries with ESAF-supported programs (Abed and others, 1998), which showed that countries that experienced a high number of interruptions of IMF-supported programs tended to have higher levels of current expenditures and lower capital outlays (relative to program targets) than countries with few or no interruptions.

When fiscal consolidations are supported by more buoyant tax revenues, the probability of ending an adjustment is lower. The results in Table 2 show that accelerated tax revenue collection increases the probability that the consolidation effort will be sustained. This result is at variance with the findings for industrial countries, where adjustments based on higher tax revenue were found to be less successful. However, in the context of low-income countries—where revenue ratios to GDP are generally modest—higher tax revenue collection can be triggered by improvements in tax administration, elimination of exemptions, and curbing of tax evasion, rather than an increase in tax rates. These factors are likely to have a positive effect both on the fiscal stance and on growth, thereby increasing the probability that an adjustment will last longer.

The availability of external financing tends to reduce the probability of continuing a fiscal consolidation in Model 1, while there is no evidence that this is true for domestic financing in Model 3. The coefficient for external financing is significant at the 5 percent level, even though including the share of either external or domestic financing in total deficit financing fails to lead to significant coefficients.

Finally, we find moderate empirical support in favor of an independent effect of economic growth on the duration of the fiscal adjustment. The probability of ending a fiscal consolidation effort is negatively related to per capita growth, as expected, but the coefficient is significantly different from zero only at the 10 percent level in Model 3.

Table 3.Sustainability of Fiscal Consolidations in Low-Income Countries by Group of Countries: Survival Function
Pre-stabilization Countries
IntervalEstimateStandard error95% confidence interval
010.9030.0160.8650.930
120.4410.0300.3810.499
230.1940.0290.1410.254
340.0610.0230.0270.116
450.000
Source: Authors’ calculations.
Post-stabilization Countries
IntervalEstimateStandard error95% confidence interval
010.9030.0350.8080.953
120.3730.0620.2540.492
230.2070.0610.1030.336
340.0690.0500.0100.209
450.000
Source: Authors’ calculations.
Source: Authors’ calculations.

These results hold for all countries in the sample, independently of the degree of macroeconomic stability achieved during the period. To test the assumption of homogenous survival functions between countries, we split the sample into two groups: post-stabilization and pre-stabilization countries. Fiscal deficits are used to identify “post-stabilization” countries.10 In Table 3, we compare the life tables for the two groups and report the survival functions. Results show that fiscal adjustments in post-stabilization countries have a slightly lower probability of survival after two periods. In fact, the survival rate is 37 percent compared with 44 percent in the pre-stabilization group. However, the difference between these two estimates cannot be considered significant as the confidence intervals for the post-stabilization groups is very large, ranging from 25 percent to 49 percent. This range includes the range of estimated survival probabilities for pre-stabilization countries (from 38 percent to 50 percent). To carry out a formal assessment of the equality of the survival functions in the two subsamples, we perform a Wilcoxon rank test. The χ2 for this test is equal to 0.92 with one degree of freedom. The level of probability associated to this value is 0.34 and does not allow rejection of the null hypothesis that the two group-specific survival curves are equal.

Table 4.Sustainability of Fiscal Consolidations and Budget Composition in Low-Income Countries: Results from Stratified Cox Proportional-Hazard Model, 1990–20001
Model 1Model 2Model 3
Coefficientz-testCoefficientz-testCoefficientz-test
Size of adjustment−0.04−3.12***−0.03−3.77***−0.043.71***
Initial deficit0.211.600.022.17**0.022.15**
Change in growth−0.02−2.17**−0.02−2.61***−0.02−2.06**
Change in social spending/GDP−0.01−1.000.010.250.020.53
Number of previous failures0.013.58***0.014.38***0.013.91***
Change in tax revenues/GDP−0.09−2.94***−0.11−5.03***−0.07−2.19**
Change in transfers/ current spending0.032.24**0.021.82*0.021.87*
Change in current/ total spending0.134.01***0.125.19***0.113.17***
Change in external financing/GDP0.074.54***
Change in domestic financing/GDP0.010.02
Number of episodes167188167
Number of failures107118107
Time at risk239272239
Log likelihood−418.3−477.8−423.1
Wald test87.3477.1367.29
Probability0.000.000.00
Pseudo-R20.220.170.17
Note: Significance at the 10, 5, and 1 percent levels is indicated by *, **, and ***, respectively.

Maximum likelihood estimates with robust standard errors.

Note: Significance at the 10, 5, and 1 percent levels is indicated by *, **, and ***, respectively.

Maximum likelihood estimates with robust standard errors.

An alternative robustness analysis to assess the effect of macro-economic stability on the duration of fiscal adjustments is based on a stratified Cox proportional hazard regression. This consists of assuming that the baseline hazard rate is allowed to be different between post-stabilization and pre-stabilization countries. Results are reported in Table 4, which shows almost identical coefficients for Models 1-3 compared to the baseline regressions in Table 2. A final Wald test has been carried out for each model to test the significance of the inclusion of the post-stabilization dummy in the baseline regression. The results of these tests do not allow rejection of the hypothesis that the effect of the post-stabilization dummy is zero.

Parametric Analysis

This section reports the results of the parametric analysis on the determinants of fiscal adjustment duration. In these models, we assumed that the hazard function has a predefined shape based on the prior knowledge of the event to be studied. In the case of the analysis of fiscal consolidation spells, the economic theory does not predict the distribution of the baseline hazard rate. The instantaneous probability of ending a fiscal consolidation may increase with time if fiscal adjustments become more costly the longer they last. However, fiscal consolidation efforts may also be self-reinforcing because the adjustment episodes based on non-permanent measures tend to be reversed sooner than deficit reduction based on structural policies. The latter tend to be a selected sub-sample of the total number of spells. Furthermore, the baseline probability of ending a fiscal consolidation may fall with time as a result of intrinsic adjustment fatigue not modeled in the covariates vector.

Since the choice of the distribution governing the duration process is unknown, it is recommended to minimize the assumptions regarding the shape of the baseline hazard function. This is in fact what we have done in the semi-parametric analysis discussed in the subsections above where the form of the hazard function was not restricted deductively. In this subsection, we compare the results of Cox’s proportional hazard model with two flexible parametric models based on the Weibul and the Gompertz distribution. The only assumption made using these distributions is that the baseline hazard function is monotonic. Whether the failure rate falls or increases with time is based on the sign of the estimated ancillary parameter. In the case of a monotonically increasing baseline hazard function, the Weibul distribution’s ancillary parameter is higher than unity, while it has to be positive in the case of the Gompertz function.

Results in Table 5 report the estimated hazard ratios based on the Cox’s proportional hazard model and the two parametric models described above. To facilitate the comparison in terms of goodness of fit between the parametric models we also calculate the AIC statistic, which shows that the model based on the Weibul distribution should be preferred to the model based on the Gompertz function. The ancillary parameters for both models point to a positively sloped, monotonically increasing baseline hazard function. This means there could be some empirical evidence that the baseline risk of ending a fiscal consolidation is increasing with time. It does not, however, mean that the total hazard is also increasing with time.

The results of the regression coefficients (hazard ratios) of the two parametric models can be compared with the baseline regression. In general, results are fairly consistent across the different models. In particular, although the distribution of the baseline hazard function was found to be positively sloped with respect to time, the results confirm that the size of the adjustment is positively correlated with the duration of the fiscal consolidation spells.

Table 5.Sustainability of Fiscal Consolidations and Budget Composition in Low-Income Countries: Results from Parametric Models, Model 1, 1990–20001
Cox Proportional ModelWeibulGompertz
Coefficientz-testCoefficientz-testCoefficientz-test
Size of adjustment0.96−3.06***0.94−3.59***0.943.57***
Initial deficit1.011.201.021.131.010.79
Change in growth0.98−1.99**0.97−1.99**0.97−1.93*
Change in social spending/GDP0.96−1.071.00−0.080.98−0.40
Number of previous failures1.013.75***1.027.74***1.006.98***
Change in tax revenues/GDP0.92−2.51**0.90−1.86**0.91−1.73*
Change in transfers/ current spending1.032.18**1.041.76*1.042.09**
Change in current/ total spending1.133.92***1.111.99**1.122.17***
Change in external financing/GDP1.074.22***1.083.22***1.083.90***
Ancillary parameter3.6518.81***1.5010.52***
Number of episodes167188167
Number of failures107118107
Time at risk239272239
Log likelihood−467.43−61.77−91.41
Wald test86.62114.77106.60
Probability0.000.000.00
AIC statistic145.54204.82
Note: Significance at the 10, 5, and 1 percent levels is indicated by *, **, and ***, respectively.

Hazard ratios. Maximum likelihood estimates with robust standard errors.

Note: Significance at the 10, 5, and 1 percent levels is indicated by *, **, and ***, respectively.

Hazard ratios. Maximum likelihood estimates with robust standard errors.

Conclusions

In this chapter, we have used survival analysis to assess the factors underlying the duration of fiscal adjustment episodes in a sample of 39 low-income countries during the period 1990-2000. The results show that tilting the overall composition of public expenditure toward more productive uses is particularly important for achieving more sustained fiscal adjustments.

Fiscal consolidations achieved by cutting selected current expenditures tend to last longer than adjustments based on cuts in more productive spending. According to the results of our analysis, protecting capital expenditures during a fiscal adjustment leads to a longer fiscal consolidation episode, as does an increase in the share of current spending on nonwage goods and services.

The size of the fiscal adjustment effort and economic growth conditions are important for the persistence of fiscal consolidations. Countries with larger cumulative reductions in the deficit are less likely to abandon their adjustment efforts than others. Furthermore, the probability of ending a fiscal consolidation effort is negatively related to per capita growth.

Our results depart from the empirical literature on industrial countries in one significant aspect. When fiscal consolidations are supported by an accelerated pace of revenue collection, the probability of ending an adjustment is lower, while expenditure reductions play a minor role. This contrasts with findings in the literature on industrial countries, where expenditure reductions dominate the sustainability of fiscal adjustment. This result has important policy implications. As revenue-to-GDP ratios are particularly low in these countries, in part because of weak administration, narrow tax bases, and tax avoidance, there is scope to mobilize revenue without raising tax rates.

Finally, we find that having achieved macroeconomic stability does significantly affect the probability of ending a fiscal consolidation. Recent studies have found that the effects of fiscal policy on growth tend to be nonlinear (Adam and Bevan, 2001; Gupta and others, 2002). However, we do not find sufficient evidence that countries that have not yet achieved stable macroeconomic conditions behave significantly differently from post-stabilization economies with regard to the duration of fiscal adjustments. Nonetheless, a country with unfavorable initial fiscal conditions is more likely to end a fiscal consolidation.

References

    AalenOlivier1978“Nonparametric Inference for Family of Counting Processes,”Annals of StatisticsVol. 6 pp. 70126.

    AbedGeorge and others1998Fiscal Reforms in Low-Income Countries: Experience Under IMF-Supported ProgramsIMF Occasional Paper No. 160 (Washington: International Monetary Fund).

    AdamChristopher and DavidBevan2001“Nonlinear Effects of Public Deficits on Growth,”paper prepared for the Cornell/ISPE Conference on Public Finance and DevelopmentSeptember7-9New York.

    AdamChristopher and DavidBevan2003“Staying the Course: Maintaining Fiscal Control in Developing Countries,”paper prepared for the Brookings Institution Trade Forum ConferenceMay15-16Washington.

    AlesinaAlberto and SilviaArdagna1998“Tales of Fiscal Adjustment,”Economic PolicyVol. 27 (October) pp. 487546.

    AlesinaAlberto and RobertoPerotti1995“Fiscal Expansions and Adjustments in OECD Countries,”Economic PolicyVol. 21 (October) pp. 20548.

    AlesinaAlberto and RobertoPerotti and JoseTavares1998“The Political Economy of Fiscal Adjustments,”Brookings Papers on Economic Activity: 1Brookings Institution.

    ArdagnaSylvia2001“Fiscal Policy Composition, Public Debt, and Economic Activity,”Public ChoiceVol. 109 (December) pp. 30125.

    ChuKe-young and others1995Unproductive Public Expenditure: A Pragmatic Approach to Policy AnalysisIMF Pamphlet No. 48 (Washington: International Monetary Fund).

    ClevesMarioWilliamGould and RobertoGutierrez2002An Introduction to Survival Analysis Using Stata (College Station, Texas: Stata Press).

    CoxD.1972“Regression Models and Life Tables,”Journal of the Royal Statistical SocietySeries BVol. 34 pp. 187220.

    GuptaSanjeevBenedictClementsEmanueleBaldacci and CarlosMulas-Granados2002“Expenditure Composition, Fiscal Adjustment, and Growth in Low-Income Countries,”IMF Working Paper No. 02/77 (Washington: International Monetary Fund).

    GuptaSanjeevBenedictClementsEmanueleBaldacci and CarlosMulas-Granados2004“Fiscal Policy, Expenditure Composition, and Growth in Low-Income Countries,”Chapter 2 in Helping Countries Develop: The Role of Fiscal Policyed. by SanjeevGuptaBenedictClements and GabrielaInchauste (Washington: International Monetary Fund). Also published in theJournal of International Money and FinanceVol. 232004.

    GuptaSanjeevBenedictClementsEmanueleBaldacci and CarlosMulas-Granadosforthcoming“The Persistence of Fiscal Adjustment in Developing Countries,”Applied Economics Letters.

    HosmerDavid W. and StanleyLemeshow1999Applied Survival Analysis: Regression Modeling of Time to Event Data (New York: John Wiley and Sons).

    IzeAlain1991“Measurement of Fiscal Performance in IMF-Supported Programs: Some Methodological Issues,” in How to Measure the Fiscal Deficit: Analytical and Methodological Issuesed. by Mario I.Blejer and AdrienneCheasty (Washington: International Monetary Fund).

    KaplanE.I. and P.Meier1958“Nonparametric Estimation from Incomplete Observations,”Journal of the American Statistical AssociationVol. 53 pp. 45781.

    MarotoReyes and CarlosMulas-Granados2001“The Duration of Fiscal Consolidations in the European Union,”Fundación de Estudios de Economia Aplicada (FEDEA) Working Paper No. 2001/19 (Madrid, Spain).

    PerottiRoberto1998“The Political Economy of Fiscal Consolidations,”Scandinavian Journal of EconomicsVol. 100No. 1 pp. 36794.

    von HagenJürgen and RolfStrauch2001“Fiscal Consolidations: Quality, Economic Conditions, and Success,”Public ChoiceVol. 109 (December) pp. 32746.

    von HagenJürgenAndrew HughesHallett and RolfStrauch2002“Budgetary Consolidation in Europe: Quality, Economic Conditions, and Persistence,”Journal of the Japanese and International EconomiesVol. 16 (December) pp. 51235.

The authors wish to thank Shamit Chakravarti and Erwin Tiongson for their help in preparing this chapter. Carlos Mulas-Granados was a summer intern in the Fiscal Affairs Department in the summer of 2001.

The countries are Albania, Armenia, Benin, Bolivia, Burkina Faso, Cambodia, Cameroon, the Central African Republic, Chad, Djibouti, Ethiopia, The Gambia, Ghana, Georgia, Guinea, Guinea-Bissau, Guyana, Honduras, Kenya, the Kyrgyz Republic, Laos, Lesotho, Macedonia (FYR), Madagascar, Malawi, Mali, Mauritania, Moldova, Mozambique, Nicaragua, Niger, Rwanda, São Tomé and Príncipe, Senegal, Tajikistan, Tanzania, Vietnam, Yemen, and Zambia.

The difference between revenues and expenditures can be different from the cash deficit in those countries that measure expenditures on a commitment basis.

This roughly corresponds to the low-deficit country group identified in Abed and others, 1998.

As a robustness check, the analysis was also conducted using an alternative threshold of ½ percentage point of GDP, with broadly similar results being obtained.

The tests are based on the comparison across groups of the expected survival under the assumption that the failure rate is the same for each group, using an appropriate weighting matrix. Under the null hypothesis, the test is distributed as a χ2 with r − 1 degrees of freedom, where r is the number of groups.

Mathematically, the baseline hazard function ho(t) is defined for all time t in which a change has taken place, and is not defined for other moments of time. But the survivor function So(t) is defined for all values of t. It follows that h(t, X) = h0(t)*exp(X′β).

This test is based on the assumption that if the model is correctly specified, the Schoenfeld residuals would not be correlated with time. Although a general goodness of fit statistic can be easily calculated as

its interpretation is not straightforward as it depends on the number of censored events in the sample. An alternative approach is based on the analysis of the residual as suggested in Hosmer and Lemeshow (1999).

Non-nested models can be compared using the Akaike Information Criterion (AIC) based on the value of the maximized log-likelihood L as follows:

where k is the number of covariates and c is the number of model-specific distribution parameters.

For example, we can assume that the sample is split into lower-income and higher-income countries, and a dummy variable having value equal to unity for the lower-income countries is used as a regressor. In this case the hazard ratio coefficient on this variable will measure the risk of ending a fiscal adjustment spell for a country that is in the lower-income group relative to the average risk in the higher-income group.

This roughly corresponds to the low-deficit country group identified in Abed and others, 1998. Based on this criterion, only seven countries can be considered post-stabilizers (Benin, The Gambia, Lesotho, Macedonia, FYR, Mauritania, Senegal, and Tanzania).

    Other Resources Citing This Publication