Public Expenditure and Consumption Volatility

Recent estimates of the welfare cost of consumption volatility find that it is significant in developing nations, where it may reach an equivalent of reducing consumption by 10 percent per year. Hence, examining the determinants of consumption volatility is of utmost relevance. Based on cross-country data for the period 1960-2005, the paper explains consumption volatility using three sets of variables: one refers to the volatility of income and the persistence of income shocks; the second set of variables refers to policy volatility, considering the volatility of public spending and the size of government; while the third set captures the ability of agents to smooth shocks, and includes the depth of the domestic financial markets as well as the degree of integration to international capital markets. To allow for potential endogenous regressors, in particular the volatility of fiscal policy and the size of government, the system is estimated using the instrumental variables method. The results indicate that, besides income volatility, the variables with the largest and most robust impact on consumption volatility are government size and the volatility of public spending. Results also show that deeper and more stable domestic financial markets reduce the volatility of consumption, and that more integrated financial markets to the international capital markets are associated with lower volatility of consumption.


Policy ReseaRch WoRking PaPeR 4633
Recent estimates of the welfare cost of consumption volatility find that it is significant in developing nations, where it may reach an equivalent of reducing consumption by 10 percent per year. Hence, examining the determinants of consumption volatility is of utmost relevance. Based on cross-country data for the period 1960-2005, the paper explains consumption volatility using three sets of variables: one refers to the volatility of income and the persistence of income shocks; the second set of variables refers to policy volatility, considering the volatility of public spending and the size of government; while the third set captures the ability of agents to smooth shocks, and includes the depth of the domestic This paper-a product of the Policy Review Unit, Development Economics Vice-Presidency-is part of a larger effort in the department to analyze the effects of public policy and its relationship with consumption volatility in developing countries. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at sherrera@worldbank.org, bruno.vincent@polytechnique.org. financial markets as well as the degree of integration to international capital markets. To allow for potential endogenous regressors, in particular the volatility of fiscal policy and the size of government, the system is estimated using the instrumental variables method. The results indicate that, besides income volatility, the variables with the largest and most robust impact on consumption volatility are government size and the volatility of public spending. Results also show that deeper and more stable domestic financial markets reduce the volatility of consumption, and that more integrated financial markets to the international capital markets are associated with lower volatility of consumption.

Introduction
Several papers have estimated the welfare cost of consumption or output volatility in developing countries, and though there is a wide range of variation in results across countries, a general finding is that the cost is significantly higher than in developed nations; the average cost in a sample of countries may reach the equivalent of reducing household consumption by 10 percent (Atahasoulis and van Wincoop, 2000;Pallage and Robes, 2003;De Ferranti, et.al. 2000;Herrera, 2007). Given this extraordinarily high cost, it seems relevant determining the causes of consumption volatility and its relationship with economic policy.
Previous papers have focused on the determinants of output volatility. In particular, Fatas and Mihov have written a series of papers on this topic (2003,2004,2005), and this one follows closely their methodology to examine the determinants of consumption volatility. The explanation of consumption volatility is based on three sets of variables: one referring to the volatility of income and the persistence of income shocks; the second set refers to policy volatility, considering the volatility of public spending and the size of government; the third set captures the ability of agents to smooth shocks, and includes the depth of the domestic financial markets as well as their integration to international capital markets.
The paper has four main sections. Section 2 presents stylized facts concerning output, consumption, and government spending across the world in the period 1960-2005. Section 3 discusses the construction of our indicator of fiscal policy volatility. Section 4 explores the relationship between fiscal policy volatility and household consumption volatility, and Section 5 presents the results of the analysis. Section 6 provides a conclusion and directions for further research. Table 1 shows the average and the standard deviation of the growth rates of GDP, household consumption and public spending in the period 1960-2005. Regarding the average growth rates, low income countries' GDP and consumption are the lowest, while their public spending grew at the highest rate. The average growth rates of consumption and GDP of middle-income and high-income countries are almost identical, but public spending growth is significantly higher in middle-income countries.

Stylized Facts of Government Expenditure, GDP and Consumption
Regarding the volatility of the three aggregates, it is the highest in the low-income counties, and decreases with the group's income. Public spending is significantly more volatile than output and consumption in low and middle-income countries. Table 2 shows the volatility of public spending is almost three times the volatility of output in HIPC and LDC, while in high income the volatility of output is higher than that of public spending. Table 2 also shows that consumption is more volatile than income in low and middle-income countries but the opposite happens in the high-income category  The fact that public spending volatility is higher than that of output and consumption in low and middle-income countries is not necessarily problematic. In fact, it could be interpreted as a successful use of a policy instrument to stabilize the target variable (output or consumption). However, the pro-cyclical nature of fiscal policy in most developing countries, explored elsewhere (Alesina and Tabellini, 2005;Talvi and Vegh, 2005), suggests that fiscal policy may be destabilizing. This is confirmed by the positive association between fiscal policy volatility and output and consumption volatility, as shown in Figures 1 and 2. 2 2 Volatility is defined as the standard deviation of the per-capita GDP and consumption growth rates during 1961-2005.

Figure 1: Fiscal Policy Volatility and GDP Volatility
Source: Calculations are based on WDI data. Volatility is defined as the standard deviation of growth rates.

Figure 2: Fiscal Policy Volatility and Household Consumption Volatility
Source: Calculations are based on WDI data. Volatility is defined as the standard deviation of growth rates The relationship between fiscal policy volatility and output volatility ( Figure 1) has been studied by Fatas and Mihov in a series of papers, but the relationship with consumption volatility ( Figure 2) has yet to be examined carefully, which is the objective of this paper. The positive association may be due to a third (omitted) variable that affects both.
Figure 3 depicts one last fact worth highlighting, namely the positive association between the average growth rate and the volatility of public spending. The positive association can be refined further by noting that there are two distinct groups of countries: the first, composed mostly of the OECD and East Asian economies, shows a positive relationship between the variables, but volatility is significantly lower for any level of growth; Chile and South Africa are included in this group. The second group, composed mostly of Latin American and African economies, also shows a positive association between the average growth rate and the volatility, though much steeper.
The "growth champions" 3 are all on the lower schedule (first group), while the poorer countries tend to be on the higher schedule, making it possible that the income level explains part of the observed relationship between volatility of public spending and consumption volatility. Hence, the remainder of this paper undertakes a more detailed examination of the determinants of consumption volatility.

Figure 3: Public Spending Growth and Volatility: 1961-2005
Source: Calculations are based on WDI data.

Measuring Fiscal Policy Volatility
The first step of our analysis is to construct an indicator of fiscal policy (or public spending) volatility that is independent of the business cycle and reflects discretionary policy changes. Following Fatas and Mihov (2003), discretionary fiscal policy is defined as the residual ε i,t of the following model: where: -ΔG is the growth rate of government spending 4 ; -ΔY is the current output growth; -W is a vector of control variables including inflation and an oil prices.
The volatility of fiscal policy is calculated as the standard deviation of ε i,t for any country i.
Equation (1) was estimated by individual country regressions and by panel methods, using both OLS and instrumental variable (IV) methods, for a total of four alternative estimation methods. Individual country regressions have the advantage of allowing heterogeneity in the estimated coefficients. 5 The panel specifications adopted in this paper impose the coefficient homogeneity assumption, but by including the level of income (GDP per capita) among the control variables we control for the potential source of bias (described in Figure 3), according to which the poorest countries are more volatile simply because of their lower income level. 6 The Instrumental Variables regressions use the following variables to instrument for current output growth: two lags of output growth; index of oil prices; lagged inflation; lagged value of government spending growth.
The estimates of policy volatility obtained by the individual country regressions (time series) and panel methods are relatively similar, and are reported in Appendix 2. Both methods yield rankings of countries with a clustering of developed economies at the top of the list (low volatility) whereas developing countries, especially those in Africa, cluster at the bottom of the list (high volatility). Appendix 2 also compares the country rankings with Fatas-Mihov results, and with a more basic estimate of volatility which is the standard deviation of the growth rate during the sample period. In general, all the rankings produce similar results.

Exploring the Relationship between Consumption Volatility and Fiscal Policy Volatility
This section explains consumption volatility with three sets of variables: the first one refers to the income generating process, considering both the volatility of income and the persistence of income shocks; the second is related to fiscal policy; while the third set of variables refers to the agents' possibilities to smooth income shocks, that depend on the financial sector depth as well as its integration to international capital markets.
The basic model that we used follows Fatas-Mihov's specification: (2) where: σ i y is the volatility of household consumption -calculated as the standard deviation of the growth of household consumption per capita between 1960-2005; σ i ε is the volatility of fiscal policy -described in the previous section; -X i is a vector of control variables that reflect mostly the volatility of income and persistence of income shocks, as well as the availability of domestic and external financing for consumption smoothing.
X i includes the following variables: -Volatility of income; -Persistence of income; -Domestic financial sector development (level and volatility of financial liabilities); -Financial openness; -Government size; -Trade variable, defined as the ratio of imports and exports to GDP.
We will now describe in more details some of the explanatory variables that we have been using in Equation (2).

Income variables: Volatility and persistence of shocks
The empirical testing of the relationship between consumption and income and their volatilities has been studied extensively (Christiano, 1987). Based on the literature, we include the volatility of income and the persistence of income shocks as explanatory variables for the volatility of household consumption. Intuitively, the higher the volatility and the persistence of income shocks, the higher the volatility of household consumption.
These two parameters are estimated with the following augmented Dickey-Fuller regression 7 : 7 All the variables are in log terms, except the time trend.
The volatility of income is defined as the standard deviation of ε t whereas the persistence parameter is defined as 1+α. Some of the specifications that we used included an interaction term between both variables.

Financial markets: Possibility of smoothing shocks
The possibility of smoothing consumption will depend on the depth and efficiency of financial markets. The relationship between household consumption volatility and financial markets has been explored recently by Bekaert, Harvey and Lundbal (2006). Here we use financial rather than equity market information, simply because of the availability of data for the sample period and the large set of countries that we are considering. We consider the ratio of total liquid liabilities, defined as the sum of currency, demand and interest-bearing liabilities of banks and other financial intermediaries, to GDP. We also include the volatility of financial liabilities.
Finally, the openness of the capital account is captured through the Chinn-Ito Financial Openness Variable, because the latter coincides with our period of analysis and is available for the entire sample of countries 8 .

Fiscal policy variables: Expenditure volatility and government size
In their study of output volatility, Fatas and Mihov included two variables to describe fiscal policy, namely the volatility of fiscal policy and government size. Fiscal policy volatility may be linked to consumption volatility through the connection between public and household budgets. Government size itself is found to be associated with higher output volatility on several occasions in the literature. Gali (1994) explains the positive relationship between government size and volatility based on the resulting higher labor supply elasticity due to the lower (after tax) labor productivity and the resulting lower employment levels 9 . A recent paper looking at OECD countries also shows that countries with higher share of government expenditure in domestic output show higher sensitivity of their fiscal position to fluctuations in the business cycle (Girouard and Andre, 2005). More recently, Angeletos and Panousi (2007) explore the supply side effects of government spending and reach a similar conclusion but their result operates through a different channel: given the negative wealth effect of higher government consumption, individuals become more risk averse, which increases the risk premium, thus discouraging investment and leading to lower employment levels. This in turn results in higher labor supply elasticity, so that government size again behaves as an "automatic destabilizer" rather than the opposite.
To control for endogenous regressors and deal with measurement error problems, we followed the Fatas-Mihov instrumental variables estimation procedure. This is of particular importance regarding the volatility of fiscal policy (Equation (1)). Indeed, as described in the first part of this paper, countries with higher volatility of consumption tend to have more volatile public spending (Table 2), so that fiscal policy volatility could be endogenous to consumption volatility.
Following Fatas and Mihov, policy volatility was instrumented with institutional characteristics of the electoral and political systems that shape fiscal policy outcomes. Such variables include: the nature of the electoral system (majoritarian vs. proportional); -the nature of the political system (presidential vs. parliamentary); -the number of elections for the executive and legislative branches; -the extent of political constraints on the executive to manipulate fiscal policy. Table 3 summarizes the results of the regressions of the volatility of fiscal policy on these four instruments, based on the volatility numbers calculated from Equation (1) using a panel specification 10 .
All the instruments -except the number of elections when it is the only regressorappear very strongly correlated with our measure of the volatility of fiscal policy. The level of the R-squared of the regression with the four instrumental variables included altogether (0.61) gives confidence in the use of these variables as instruments for the volatility of fiscal policy. The other variable that captures fiscal policy, namely the size of government, was also instrumented using the trade ratio, the dependency ratio, the urbanization rate, the log of total population, and the log of GDP per capita 11 . A key difference with the Fatas-Mihov paper consists in using trade as an instrument for the size of government. In general, Fatas and Mihov estimations include simultaneously both openness and government size, leading to non-robust statistical results on these coefficients. Indeed, following Rodrik (1998), a more open economy increases the vulnerability of domestic agents and hence governments will have more pressure to spend. Hence, in some of our specifications, the size of government was instrumented with the trade variable.

Consumption Volatility and Fiscal Policy: Results
Initially a cross-section of countries is used, using the average of the variables for the period 1960-2005. In order to exploit the time dimension of the data, the sample was then split in three 15-year sub-periods, and the model was estimated by panel methods.

Cross-country results
The cross-country regressions used two alternative measures of the volatility of fiscal policy: the individual time-series regressions estimates, and the panel estimates. Results are very similar: panel estimates are reported in Table 4, while time-series specifications are mentioned in Appendix 3. Table 4 below shows the regression results, and is divided in three panels: the first one, labeled OLS, covers the 4 first columns and refers to basic Ordinary Least Squares regressions with various controls. The second panel, labeled IV-1, covers columns (5) to (8) and refers to Instrumental Variables regressions, using instruments for the volatility of fiscal policy. The third panel, labeled IV-2, covers columns (9) to (11) and reports the results of three IV regressions instrumenting both the volatility of fiscal policy and government size.
The main results of our analysis are as follows: 1-Whatever method is used and whatever control variables are included, the fiscal policy volatility is always highly significant, as least at the 90% level of confidence 12 , with a positive sign. This is a confirmation of the very strong (positive) and significant impact of the volatility of public spending on the volatility of consumption. Despite the high number of specifications that are considered, the parameter estimates for the volatility of fiscal policy are similar, clustering in the 0.3-0.5 interval, apart from the regressions where no additional controls are included.

2-
The volatility of income is always significant -at the 99% level of confidence -with a strong (positive) parameter estimate, whereas the persistence of shocks is never so.

3-
The level of financial development, as measured by the ratio of liabilities to GDP is not significant, but the volatility of the financial liabilities is always significant, at least at the 95% level of confidence, with the expected sign (positive). Capital account openness is always significant, at least at the 90% level of confidence, again with the expected sign (negative).

4-
The parameter estimates and the level of significance of the trade and government size variables vary considerably. NB: The p-values in the parentheses are based on heteroscedasticity-robust standard errors. In the IV regressions, the over-identifying test (OID) reports the pvalue from the Sargan-Hansen J-test that the instruments are uncorrelated with the residuals. All regressions include an intercept.

Panel results
In order to exploit the time dimension of our sample, we also estimated model (2) by panel regression, splitting the sample in three 15-year sub-periods: 1960-75, 1975-90, and 1990-2005. The choice of these three sub-periods is arbitrary, but is similar to Wolf's (2004) decadal analysis.
The definition of the variables was the following: The volatilities of fiscal policy and income were defined, for each country and each sub-period, using the same panel regressions as in the previous parti.e.
Equations (1) and (3) -as the standard deviation of the corresponding residual for each country and each sub-period. This was not possible for the persistence of income -that was calculated using country time-series regressions -so we used the 1960-2005 average of this parameter for the 3 sub-periods Most of the institutional variables that we used for instrumenting the volatility of fiscal policy were not available for the first sub-period. Indeed, three of the four variables that we are using -nature of the political system, nature of the electoral system and number of elections -were only available from 1975 onwards. Therefore, we used the 1960-2005 average for the four instruments of fiscal policy volatility 13 .
These adjustments to the data have a cost in terms of the variety of specifications that we can use. Indeed, as long as a variable is assumed to be constant throughout the whole period 1960-2005, it will be dropped in any panel regression using fixed effects. This is the case for the instrumented volatility of fiscal policy -since some instruments are constant during the whole period due to lack of data -as well as the persistence of income. Therefore, in order to check the robustness of our results to the instrumentation of the volatility of fiscal policy, we did not include any fixed effects in our regressions. Instead, we used a basic pooling.
The results of the panel regressions are shown in Table 5. The first 4 columns are fixedeffects panel OLS regressions. The next group of columns (IV-1) is the basic pooling with the instrumentation of fiscal policy volatility, and the last set of columns (IV-2) includes the instrumentation of government size as well. Table 5 shows that results are very similar to the cross-country estimations reported in Table 4. All the variables have the expected signs and there are no major changes in the statistical significance of the coefficients by shifting from OLS to instrumental variable (IV) estimations. However, the magnitude of some of the coefficients varies significantly. For instance, the coefficient of the volatility of fiscal policy almost doubles as well as the coefficient of the volatility of financial liabilities. Some differences with the cross-country results are worth noting, however: 1-The persistence of income indicator turns significant in two of the panel specifications -using the income variables as controls, but excluding the financial variables.

2-
The level of financial liabilities is always significant at the 90% level of significance -contrary to the cross-country regressions. The negative sign is as expected, and relates to the capacity of financial markets to smooth consumption.

3-
In the panel regressions with fixed effects, the volatility of financial liabilities is not significant anymore -just the level of financial liabilities seems to matter.

4-
P-values of the over-identifying tests are, in general, much lower than in the crosscountry regressions. This would mean that our instrumentation works better on the panel sample than on the cross-section sample.

Conclusions and Directions for Future Work
The volatility of household consumption is mostly explained by the volatility of income and that of fiscal policy. Government size, as measured by public expenditure as a share of GDP, also affects positively consumption volatility. The policy implications are clear: rules that stabilize public spending will also stabilize private consumption, with the corresponding welfare gain. The impact of public spending volatility on private consumption volatility has implications for program or project evaluation: unless additional public spending is funded with spending cuts in other areas, it will imply higher private consumption volatility, with its implicit welfare cost.
Additional work is needed to better understand this linkage. Whether public spending volatility affects private consumption volatility because of labor supply effects -as in Gali (1994) -or through its impact on risk aversion and investment -as in Angeletos and Panousi (2007) -still needs to be verified in future micro research.
These results provide support for policies that promote financial deepening and foster the development of capital markets that stabilize the growth of financial liabilities. An open capital account also leads to lower volatility of household consumption.