Effects of Fiscal Consolidations in Latin America

Abstract

We use data on structural primary balances for Latin America to estimate effects of fiscal consolidations. Identification relies on a doubly robust estimator that controls for the endogeneity of fiscal policy using inverse probability weighting. Results suggest a negative multiplier on impact, with the economy starting to recover from the second year on and expanding after 5 years. Revenue-based adjustments are more contractionary on impact than expenditure-based adjustments. Among demand components, we find consumption to be generally less responsive to consolidations than investment. Results also depend on the initial levels of debt and taxation. In comparison with OECD evidence, Latin American countries display larger, but stop-and-go consolidations, with state-dependent long-run multipliers.

Appendix

1.1 Four Definitions of Fiscal Episodes

The baseline results were presented according to a definition of fiscal adjustment that is, itself, based on the episodes that overlap between four subdefinitions. We here present details on their construction.

Denoting $\mathrm{AA}_{i,t}$ the occurrence of a fiscal consolidation in country i at period t and $\mathrm{dSBB}_{i,t}$ the change in the structural primary balance in terms of GDP, our first and second definitions are as follows:

$$\begin{aligned} \mathrm{AA}_{i,t} = \left\{ \begin{array}{l l} 1 \qquad \text {if} \qquad \mathrm{dSBB}_{i,t} \ge X \\ 1 \qquad \text {if} \qquad \mathrm{dSBB}_{i,t} \ge Y \qquad \& \qquad \mathrm{dSBB}_{t-1} \ge Y \\ 0 \qquad \text {otherwise}\\ \end{array} \right. \end{aligned}$$

This means that a fiscal adjustment is considered to be happening in a specific year if either: (i) the change in the structural primary balance-to-GDP ratio is greater or equal to X percentage points (a positive change implies a fiscal contraction since revenues minus expenditures increase) or (ii) there is a 2-year consecutive improvement in structural balance of at least Y points per year. We consider two possibilities for X and Y: $(X_1,Y_1) = (2,1.5)$ and $(X_2,Y_2) = (1.5,1.25)$, both expressed as percentage points in terms of GDP. The former resembles Alesina and Ardagna (1998), while the latter is less stringent and was chosen in order to increase the number of observations. We henceforth refer to them as AA1 and AA2.

Alternatively, the third and fourth definitions consider a fiscal episode in terms of the change in SBB relative to the standard deviation of changes along the sample. We consider both changes in structural balances relative to the whole sample standard deviation and the within-country standard deviation. We call those episodes, respectively, SD1 and SD2.

$$\begin{aligned} SD_{i,t} = \left\{ \begin{array}{l l} 1 \qquad \text {if} \qquad \mathrm{dSBB}_t \ge \sigma \\ 1 \qquad \text {if} \qquad \sum _{k=0}^{1} \mathrm{dSBB}_{i,t-k} \ge 2 \sigma \\ 0 \qquad \text {otherwise}\\ \end{array} \right. \end{aligned}$$

Table 9 presents a summary of fiscal episodes identified by each definition. One important feature of the episodes is a stop-and-go characteristic, given that very few episodes last for more than 2 consecutive years.

Episodes defined as above have a high amount of overlapping between themselves. Comparing the first and second definitions (AA1 and AA2), the latter allows for more episodes since it lowers the threshold and trivially also includes all definitions from the first one. Insofar as the third definition (SD1) considers the whole sample standard deviation, which is higher than country-specific deviation for 13 countries of the sample, a comparison between third and fourth sets of episodes is not so direct. Countries with higher than sample average dSBB standard deviation will tend to lose episodes when country-specific variations are considered, while countries with lower than average dSBB might increase their number of episodes. As seen from the table, the fourth definition has 8 episodes more, but not necessarily all episodes overlap the ones based on the third definition. The third definition is also the only one for which all countries have at least one fiscal episode. Episodes as a proportion of total observations vary between 10 and 16$\%$.

Table 9 $\#$ of fiscal episodes and observations, by definition

Table 9 also shows average change in SBB per definition, as well as initial Debt-to-GDP levels and accumulated change in debt up to 3 years prior to consolidation. One can observe that change in SBB is monotonically decreasing with the size of the consolidation treatment group. Also, the more stringent the definition, the higher the initial debt level. On the other hand, accumulated debt prior to adjustment is much higher for the last definition. Table 10 details for each country the identified episodes by definition.

Table 10 Fiscal consolidation episodes according to definition

1.2 Quality of SBBs Data: A Small Narrative-Based Assessment

A comprehensive narrative approach along the lines of the IMF approach for OECD countries is an interesting tool to provide a benchmark analysis of fiscal consolidations based on the announced fiscal measures from the budget readings.

In this section, I give an example of such a narrative description of adjustments, analyzing the episodes identified for Trinidad and Tobago, Mexico and Venezuela in 2009. It is a well-known fact that oil-exporting countries suffered greatly as a result of the Great Recession, damaging revenues in terms of both decreases in volume and prices of exports. Hence, it is highly counter-intuitive to find episodes of consolidations for those countries during this period.

Reports (see, e.g., Economic Commission for Latin America and the Caribbean 2010) and data provide interesting figures that help understand the adjustment provided by structural balances. The observation that some consolidations occur during commodity price drops is actually natural based on the definition of the SBBs. If such drops are large enough in comparison with long-run prices, even though primary balances might worsen substantially, structural balances should indeed be improved if the country is taking austerity measures in other revenue or spending items. In what follows, we briefly describe what happened in each of those countries in 2009.

• Mexico

  Public revenues in Mexico declined as the economy slowed. Oil revenues decreased due to lower prices and volumes of exports, and some nonoil tax revenues also fell due to depressed demand. This decline was not totally offset, although there were important rises in nonoil revenues and tax revenues, mainly from transfers of profits from the Central Bank and the oil revenue stabilization fund. In addition, the government cut projected nominal expenditures aiming at maintaining a stable fiscal deficit.

  The (unadjusted) primary result was negative, indicating a slight reduction in the deficit in 2009 from 2008. Tax revenues increased already in unadjusted terms. Moreover, after adjustment of taxes for cycle elasticities, tax-based revenues improved further. Hence, it seems plausible after adjusting oil revenues for their respective elasticity (very low prices compared to the “long-run” measure) to observe improvements in structural primary balances, sufficient to classify this period as a consolidation.

• Venezuela

  In Venezuela, the fiscal situation deteriorated due to a drop in government revenues (a decrease in the volume of oil exported and a decline in prices of petroleum). But as in Mexico, some efforts toward increasing tax revenues were conducted, mainly by raising VAT tax rates from 9 to 12.

  Current expenditures were increased, like public sector wages and transfers, but capital expenditures were contained. This item receives an elasticity of zero in the IDB method and hence contributes to worsening the fiscal balance, even though some cyclical adjustment could indicate a better figure.

  In net terms, the VAT tax increases, boosted by the adjustment for cycle elasticities, plus the long-run adjustment for oil revenues (as in Mexico), generate a significant improvement in the structural budget for 2009, especially when compared to the high deficit from 2008. Therefore, it also sounds plausible to consider the fiscal stance as contractionary.

• Trinidad and Tobago

  Trinidad and Tobago were probably the most affected by decreases in revenues from oil among the three countries. The decrease in international oil and gas prices exceeded forecasts used in the adjusted budget estimates from the government.

  In this case, however, the country has not adopted consolidation policies, and hence, the large positive change in (structural) balances was driven by the commodity cycle and the adjustment for price elasticities, which suggests some degree of imperfection in the measure of structural budget balances.

This small narrative example implies two things: (i) First, in 2008, negative structural balances indicate counter-cyclical policies in most economies. In 2009, the budget figure has improved, both because the previous value was very low and because some consolidation effort has been initiated by Mexico and Venezuela, in addition to the long-run adjustment for commodity prices that improves the budget when moving from observed to structural primary balances. (ii) There might be some flaws in the measurement of the fiscal stance by means of the structural balances given that Trinidad and Tobago did not implement significant contractionary measures from 2008 to 2009, but there is a big positive change in the structural balance. This possible weakness is addressed using the two-stage methodology. Likewise, in one of the specifications, we exclude those oil-dependent economies from the sample to double check if this meaningfully changes the effects of baseline estimates.

1.3 Commodity Price Indexes

In the propensity score estimation, we have included two alternative measures of commodity prices, one based on country-specific indices and the other—used as the baseline—based on the main export commodity of the country.

The first measure is a country-specific index based on five major aggregation of commodities: food, beverages, metals, oil and agriculture. The construction of the index is a simplification of the IMF methodology and uses data on specific commodity prices indices. For aggregation of commodity exports into one of these categories, we use UN Comtrade data. For each country i and period t, commodity price indices are constructed according to the following expression:

$$\begin{aligned} \Delta \mathrm{ln} (\mathrm{CommPrice})_{i,t} = \sum _{j=1}^{J} \Delta P_{j,t} \left[ \left( x_{i,j,t-1} - m_{i,j,t-1} \right) / \mathrm{GDP}_{i,t-1} \right] \end{aligned}$$

where P is the price index of each commodity category j, x are exports and m are imports.

The second measure is more simple and aims at increasing the number of observations per country, due to data limitations on the sample of imports and exports. It hence considered the country-specific commodity price index as the price index for the main commodity category in the net exports of each country.

It is apparent from Fig. 8 that this approximation works very well for the main commodity exporters of our sample. The plots display the two measures of commodity indices described. For the case of oil and metal exporters, the country-specific indices closely follow the broader “main commodity” index.

Fig. 8

Commodity price indices

1.4 Distribution of Propensity Scores, Overlapping Assumption

See Fig. 9.

Fig. 9

Distributions of propensity scores for control and treated groups

1.5 Robustness Exercises

Table 11 displays results for the four definitions compared to the baseline. One can observe a similar pattern of impulse responses across all of them.

Table 11 AIPW per 1$\%$ consolidation, different betas, no country FE

Table  12 displays results allowing different slopes for regressors and not including country FE in the conditional mean equations. In this case, the absence of country fixed-effects shifts the estimates slightly up, implying lower recessions at the beginning and faster output recovery.

Table 12 AIPW per 1$\%$ consolidation, different betas, no country FE

1.6 Data Description

See Table 13.

Table 13 Data description

1.7 Allocation Bias

See Fig. 10.

Fig. 10

OLS and AIPW estimates, booms and slumps