GROWTH VOLATILITY AND GOVERNMENT EXPENDITURE IN LOW AND MIDDLE INCOME COUNTRIES: A DYNAMIC PANEL ANALYSIS

Abstract We examine the relation between the government consumption expenditure and output growth volatility in 57 low and middle income countries by using both static and dynamic panel methods. It seems that the results of these methods largely differ from each other. Contrary to some previous results reported in the literature, we present a strong evidence for a negative relation between government expenditure and volatility in low and middle income countries. We also conclude that the volatilities of government consumption, trade openness and investment are signifi cant in explaining the growth volatility. To have a more stable economy, policy makers in these countries should pay more attention to some issues. In this context, we think that a change in the tax and expenditure system in order to make automatic stabilizers work better would be helpful. Additionally, it is important to have a sound fi scal and monetary position to effectively carry out countercyclical policies when needed. Moreover, adopting and implementing clear and fl exible rule-based economic policies should be considered. Finally, improving the institutional structure and policy making capacity must be an ultimate aim to reduce the economic volatility.


Introduction
Since a stable economy with predictable future provides a bett er business environment for economic growth, investment and decision making, ensuring macroeconomic stability is one of the most important targets for policy makers.On the other hand, higher output volatility may lead to a signifi cant cost in terms of social welfare.In this regard, several studies, such as Ramey and Ramey (1995), Fatas (2002), Hnatkovska and Loayza (2003), Badinger (2010), suggest that volatility exerts a negative eff ect on growth.Therefore, ceteris paribus, it is important to have a stable or less volatile economy.
There is an increase in the number of studies examining the output volatility especially after the 1990s, possibly motivated by the so called Great Moderation.Many studies, such as Kim and Nelson (1999), McConnell and Perez-Quiros (2000), Stock and Watson (2002), Prasad et al. (2007), and Debrun, Pisani-Ferry and Sapir (2008), show that the output volatility declined in the 1980s and 1990s in comparison to previous decades.This decline has led to a new avenue for research which focuses on the main sources or determinants of the volatility.Many factors are put forward to explain the volatility including openness, fi nancial development, macroeconomic policy, government size, and institutional structure.
Moreover, as suggested in some studies such as Kose, Prasad and Terrones (2005), Hakura (2007), Perry (2009), the output volatility has been higher in developing countries implying that there is a possibility or more space to gain by reducing output volatility in these countries.This additional gain from having a less volatile output would be more vital for developing or emerging economies that often suff er from the lack of resource, investment and predictable economic activity.Policy makers in these countries need to fi nd out the possible determinants of the volatility in order to implement some measures to have a more stable economy.Therefore, it would be particularly helpful to examine the output volatility in developing and emerging countries.
We examine the output growth volatility and the government consumption expenditure in 57 low and middle income countries by means of both static and dynamic panel estimation methods.This paper contributes to the existing literature by presenting a strong evidence for a negative relation between government expenditure and growth volatility in these countries and also employing the dynamic panel methods in this context.We also fi nd that there exists a robust and signifi cant impact of volatility of investment, government consumption, and trade openness on the output growth volatility.
The remaining of the paper is organized as follows.We summarize the literature in section 2, explain the model, data, and methodology in section 3, present and discuss the estimation results in section 4 and fi nally conclude in section 5.

Literature review
A large literature has aimed to reveal the driving factors of volatility for a long time.At the onset, we would like to point out that it is common in the literature to focus on some determinants of the volatility, largely ignoring other possible determinants.In this context, trade openness, capital fl ows and international fi nancial integration, the conduct of monetary policy, fi nancial development or deepening, fi scal variables, institutional features have been examined.Although our main motivation is to empirically examine the relationship between the government size and volatility, we also briefl y review some other key determinants.
Some arguments about a negative relationship between the government size and volatility are explained and discussed in many studies, such as Gali (1994), Fatas and Mihov (2001), Andres, Domenech and Fatas (2008), Debrun, Pisani-Ferry and Sapir (2008), Mohanty and Zampolli (2009), Debrun and Kapoor (2010).One of the arguments, which in the literature is named as composition eff ect, assumes that the public sector is more stable in general.Another argument is based on the idea that the government size would refl ect the extent or the eff ectiveness of automatic stabilizers in the economy.According to these two arguments, a large government share in the economy might contribute to reduce the output volatility or variability.
In a seminal study, Gali (1994), developing a model in which fi scal policy has a stabilizing eff ect on the output volatility, reports a negative and robust relationship between the government size and output volatility for OECD countries.This fi nding has been confi rmed by many following studies, like Fatas and Mihov (2001), Andres, Domenech and Fatas (2008), Mohanty and Zampolli (2009) for OECD countries.Martinez-Mongay and Sekkat (2005) also conclude that the relationship is not linear.Debrun, Pisani-Ferry and Sapir (2008) fi nd a negative relation between the government size and volatility but also detect that this relation has changed over time.Koskela and Viren (2004) and Viren (2005), for a large sample of countries, conclude that there is no robust evidence for a negative relationship in general.Thornton (2010) presents a strong evidence for a positive relationship for 21 Emerging Market Economies.Debrun and Kapoor (2010) report that the negative relation is signifi cant especially for OECD countries and conclude that automatic stabilizers are important.Turan (2016) doesn't fi nd any robust relation between the government size and volatility for Turkey.In a diff erent strand of the literature, Fatas and Mihov (2003) for 91 countries, Hakura (2007) for 81 countries, and Badinger (2009) for OECD countries conclude that the discretionary fi scal policy can have a positive eff ect on volatility.
Because of the close connection between openness and volatility, many studies include trade openness as a control variable.In theory, it is not clear whether openness leads to less or more output volatility (Easterly, Islam and Stiglitz , 2001).On the one hand, trade openness can enable countries to alleviate the eff ects of domestic shocks on the economy.On the other hand, it is possible that trade openness can cause an increase in the volatility by exposing the country to external shocks.Kose, Prasad and Terrones (2005) suggest that the eff ect of trade and fi nancial openness on volatility depends on some factors as the sources of shocks.Although Raddatz (2007) argues that the external shocks play only a minor role in explaining the output volatility in developing countries, Mackowiak (2007) indicates the importance of external shocks for emerging markets.Razin and Rose (1994), for 130 countries, do not fi nd any strong evidence for a relation between openness and volatility.Easterly, Islam and Stiglitz (2001), and Giovanni and Levchenko (2008), using data for a large number of countries, fi nd a positive relation between trade openness and output volatility.Calderon, Loayza and Schmidt-Hebbel (2005) present some evidence that the trade openness tends to increase the output volatility in only middle income countries.Bejan (2006), using data for 111 countries, reports a positive relation between trade openness and volatility in general and also highlights the importance of making a distinction between developed and developing countries.Cavallo (2007), for 21 OECD and 56 non-OECD countries, concludes that there exists a negative relation between trade openness and volatility.Haddad et al. (2012), utilizing data for 77 developing and developed countries, argue that the nature of the relationship between trade openness and volatility depends on the diversifi cation of a country's export base.
Financial openness, like trade openness, can help to smooth consumption and investment but also make the country more vulnerable to external shocks, sudden stops or capital outfl ows.As for fi nancial openness, Easterly, Islam and Stiglitz (2001) do not fi nd a signifi cant role for private capital fl ows with regard to volatility.Likewise Buch, Doepke and Pierdzioch (2005) examine the relationship between fi nancial openness and volatility for OECD countries and conclude that no robust relationship between these variables exists.Calderon, Loayza and Schmidt-Hebbel (2005), using data for 76 countries, do not fi nd a strong evidence that fi nancial openness leads to a greater output volatility.Bekaert, Harvey and Lundblad (2006), for a large sample of countries, suggest that fi nancial liberalization is associated with lower consumption growth volatility and do not observe any signifi cant increase in the volatility for liberalized emerging countries.Prasad et al. (2007), examining 67 industrial and developing countries, fi nd that international fi nancial integration does not help developing countries to reduce the macroeconomic volatility.On the contrary, they argue that some countries experience higher consumption volatility as a result of the fi nancial globalization.More recently Hwang, Park and Shin (2013), for 21 advanced and 81 developing countries, report that the capital market openness leads to an increase in the output volatility in developing countries.
According to the monetarist approach the problems with the conduct of monetary policy can induce some fl uctuations in the economic activity.Thus a well-implemented or high quality monetary policy can contribute to a reduction in volatility.Taylor (2000) suggests that the main reason behind the decline in the US output volatility is the bett er conduct of monetary policy after 1980s.Blanchard and Simon (2001) fi nd that there is a close association between infl ation and output volatility for the US.Debrun, Pisani-Ferry and Sapir (2008) suggest that the quality of monetary policy has a stabilizing eff ect on volatility for OECD countries.Debrun and Kapoor (2010) conclude that the central bank independence has a relation with lower volatility in only some specifi cations by utilizing data for OECD and developing countries.
As suggested in some studies, like Debrun, Pisani-Ferry and Sapir (2008), fi nancial development or deepening may be important in explaining the economic volatility.In essence, fi nancial development allows households and fi rms to smooth economic ac-tivities over time by facilitating the access to credit.It is expected that a bett er and more eff ective credit allocation mechanism could lead to a lower volatility.Easterly, Islam and Stiglitz (2001) fi nd that fi nancial variables have a stabilizing eff ect and also suggest that private credit above some level can have a positive eff ect on volatility implying a non-linear relation.Debrun and Kapoor (2010) also report a stabilizing and statistically signifi cant impact of fi nancial development on the output volatility.In a recent paper, Wang, Wen and Xu (2013) developing a model in which fi nancial development has a negative relation with volatility, conclude that there is a negative relationship between the fi nancial development and volatility in a large sample of countries.
Although the importance of institutions for economic analysis and outcomes has been recognized for a long time, it has increasingly drawn more att ention in recent years.In a comprehensive and infl uential study, Acemoglu et al. (2003) highlight the role of institutions in explaining volatility.Unlike many studies in the literature, they conclude that when the institutional structure has been taken into consideration, macroeconomic variables, including government size, have a small impact on volatility.Acemoglu et al. (2003) suggest that the institutional problems, not distortionary macroeconomic policies, are the underlying reason of observed economic volatility.Loayza et al. (2007) also mention the role of institutions for supporting policies that can reduce the volatility.However, the importance of institutions does not necessarily mean that macroeconomic policies are not eff ective or useless.On the other hand, we should note that good institutions and public administration are crucial to make and implement any policy decision in an eff ective way.
To sum up, in this wide and expanding literature, some factors that come into prominence are fi nancial development, government size, trade openness, infl ation, fi nancial openness or capital fl ows and institutions.Government size has a stabilizing eff ect on volatility only in developed countries although this eff ect has weakened over time.However, government size does not seem to be related to lower output volatility in developing countries.It seems that we can conclude that the fi nancial development, to some extent, and a good monetary policy have a stabilizing eff ect on output volatility.The eff ect of trade openness on volatility is not clear, while we can say that fi nancial openness does not appear to have a robust stabilizing eff ect.

Model, data, and empirical methodology
In order to identify the determinants of output volatility, we use the following model fi rst: where Vol and Av stand for the standard deviation and average of related variables, estimated over non-overlapping fi ve-year periods, respectively.In other words, following the literature we use the standard deviation as a proxy for volatility in this study.The dependent variable is the standard deviation of real GDP growth rate (GR).GE represents the government consumption expenditure.It would be bett er to use the general government expenditures, including especially transfer payments, rather than consumption expenditures, but, unfortunately, these series are not long enough to perform robust econometric analysis.Although our main interest is to fi nd whether there exists a negative relation between the government expenditure and output growth volatility in low and middle income countries, we control some variables, shown by X, including the trade openness (TO), domestic credit provided by the fi nancial sector (CR), investment (IV), fi nancial account balance (FA), infl ation rate (IN), and average growth rate (GR).
The trade openness (TO), measured as the sum of export and import, is widely used in the literature in examining the link between the government expenditure and volatility.We use the domestic credit provided by the fi nancial sector (CR) as a proxy for fi nancial development.Similarly, the infl ation rate (IR) is expected to capture the conduct of monetary policy.The role of investment (IN), measured as the gross fi xed capital formation, in economic fl uctuations has been widely known and recognized among economists, since at least the emergence of Keynesian theory, if not earlier.
The potential importance of capital fl ows is obvious in examining the volatility for low and middle income countries.However, due to data limitations, we use the financial account balance (FA) rather than the private capital fl ows.Finally, we also include the growth rate (GR) as an independent variable.Except for the infl ation rate and the growth rate, all independent variables are percent of nominal GDP.
It is clear that Model 1 relates the average values of independent variables to the dependent variable.However, it is possible that the volatilities of the independent variables, opposed to average values, would be the main drivers of the growth volatility.For example, Gali (1994) uses the government size in level and also its standard deviation.In a similar way, Easterly, Islam and Stiglitz (2001) employ some variables in levels and their standard deviations as well in order to identify the determinants of the volatility.In another important study, Blanchard and Simon (2001) examine the link between the output volatility and infl ation but also the output volatility and infl ation volatility.Therefore, we employ the following model which includes the volatilities instead of average values of independent variables: Since Models 1 and 2 include only average values and volatilities of independent variables, respectively, these models might be subject to omitt ed variable bias.Because both the average values and volatilities of related variables would aff ect the growth volatility at the same time, in order to obtain more robust results, we also build up the following model which includes both average values and volatilities in the same specifi cation: The models are estimated by using unbalanced panel data for 57 low and middle income countries over the period 1965-2014.We have derived the series by using data from World Development Indicators (WDI) of the World Bank (2016) with the exception of fi nancial account balance provided from Balance of Payment Statistics (BPS) of International Monetary Fund (2016).The description of the variables is summarized in Table 1.As for the methodology, we perform panel data analysis which provides the opportunity of simultaneously dealing with cross sectional units (i=1, …, N) and time dimension (t=1, …, T) in the framework of equation 4 as below: Here, it y stands for the dependent variable which is the standard deviation of real GDP growth rate as specifi ed in Models 1, 2, and 3. On the other hand, it x indicates our main explanatory variable, namely government consumption expenditure, which has 2 diff erent specifi cations with regard to the models presented above.Also, we utilize number of control variables denoted by controls which are listed above in the course of defi ning the models.As the fi rst step of the methodology, we apply 2 alternative static panel model specifi cations, that are fi xed eff ect (FE) and random eff ect (RE) models.Fixed eff ect model eliminates time-invariant diff erences between countries to control for unobserved heterogeneity.On the other hand, in the random eff ect model, unlike the fi xed eff ect model, it is assumed that the variations across countries are random and not correlated with the regressors.As Greene (2008) emphasizes, the crucial distinction between fi xed and random eff ects is whether the unobserved country specifi c eff ects embody elements which are correlated with the regressor of the models.In this regard, we select the appropriate one for modelling by means of Hausman test which has the null hypothesis of no correlation between the explanatory variables and error terms.The rejection of the null hypothesis gives support for the fi xed eff ect model.Considering the defi ciencies of static panel data analysis, we also employ dynamic panel data approach which eliminates both the cases of unobservable factors correlated with the dependent variable and regressors, and the dependent variable aff ecting the explanatory variables.The GMM panel estimator should strictly be employed when the time dimension is relatively short according to cross section units (N T) since small T is more likely to imply the endogeneity of the dependent variable.Thus, to control the potential endogeneity occured as a consequence of correlation between independent variable and error term and eliminate the unobserved individual eff ects which could potentially lead to biased and inconsistent parameter estimates, we follow the literature dealing with growth issues by applying general method of moments (GMM) approach.As an extension of GMM methodology, Arellano and Bond (1991) propose to eliminate individual fi xed eff ects by diff erencing.The diff erence GMM (DGMM) of Arellano and Bond (1991) is given below in equation 5: In addition, the extended version of the GMM estimators, that is system GMM (SGMM) suggested by Blundell and Bond (1998), is derived from the estimation of a system of two simultaneous equations which are in levels and in fi rst diff erences.We also report the results of system GMM with regard to its several priorities to diff erence GMM of Arellano and Bond (1991).Firstly, the weak instrument problem for diff erence GMM estimators in the existence of random-walk variables makes the sys-GMM estimation more preferable.Secondly, system GMM estimators are more consistent in the case of continious instrumental variables, and there is a dramatic reduction in the fi nite sample bias due to the utilization of extra moment conditions.In addition, system GMM is more appropriate when dealing with unbalanced panel due to the magnifying gaps in diff erence GMM estimation (Blundell and Bond, 1998;Blundell and Bond, 2000;Bond, 2002;Coban and Topcu, 2013).
Finally, as suggested by Arellano and Bond (1991), Arellano and Bover (1995) and Blundell and Bond (1998), we examine the consistency of GMM estimators by means of specifi cation tests which analyze the serial correlation properties of error terms and validity of instruments.For this purpose, we fi rst utilize Arellano-Bond test of autocorrelation in the fi rst diff erenced errors at order 2.Then, we apply Sargan test of overidentifying restrictions which examines the suitability of the used tools.Rejection of both null hypothesis of autocorrelation and Sargan tests avoids the problem of model misspecifi cation.

Estimation results and discussion
The results of Model 1 are presented in Table 2.It seems that our static and dynamic panel estimates largely diff er from each other.FE and RE estimates suggest that the average infl ation rate has a positive and statistically signifi cant impact on the volatility.As expected, it seems that an increase in the infl ation rate causes a rise in the volatility.This positive eff ect, consistent with theoretical explanations in the literature, like Taylor (2000), and Blanchard and Simon (2001), can be interpreted as the signifi cance of monetary policy regarding the volatility.Moreover, FE estimates indicate that the average trade openness has a signifi cant negative eff ect as well.This stabilizing eff ect of the trade openness is consistent with the results of studies that fi nd a negative relationship between the openness and volatility.On the other hand, RE results suggest the negative eff ect of the domestic credit provided by the fi nancial sector, which is in line with theoretical predictions.Since Hausman test suggests that RE estimates are more appropriate, we conclude that the average infl ation rate and domestic credit are important variables in the linear model estimates.
DGMM and SGMM results also confi rm the positive and signifi cant eff ect of the infl ation rate on the volatility.It seems that all the estimation methods we have carried out suggest the importance of infl ation rate or monetary policy for the volatility.Furthermore, DGMM results also suggest a negative and signifi cant eff ect of the government consumption expenditure on the volatility, implying that the negative link between the government expenditure and volatility is not limited to developed or OECD countries, but also valid for low and middle income countries.This fi nding diff ers from that of some studies, such as Koskela and Viren (2004), Viren (2005), and Thornton (2010), which do not report any strong negative relation between the government expenditure or size and volatility in developing countries.
We expect the volatilities of all explanatory variables to have a positive eff ect on the output growth volatility.This means that an increase in the volatilities of independent variables will lead to a rise in the growth volatility.As presented in Table 3, both FE and RE estimates indicate that only the volatilities of government consumption and investment have a signifi cant and positive eff ect.It is not surprising to fi nd that an increase in the government consumption or investment volatility would change the output growth volatility.FE results also indicate that the volatility of infl ation rate and domestic credit have also a signifi cant impact on the output growth volatility.Based on Hausman test, FE estimates should be used when interpreting the results for Model 2. Therefore, we conclude that the volatilities of government consumption, investment, infl ation rate and domestic credit are important variables in explaining the output growth volatility in the static estimates.
On the other hand, dynamic panel estimates suggest that an increase in the volatilities of trade openness and investment leads to a statistically signifi cant rise in the output growth volatility.Diff erence GMM estimates also suggest that the volatility of government consumption has a signifi cant eff ect, which is consistent with the linear models.However, unlike fi xed and random eff ect results, we do not fi nd any significant impact of the volatility in the domestic credit and infl ation rate on the output growth volatility in dynamic panel estimates.It seems that the eff ects of volatilities of government consumption and investment are robust across diff erent estimation methods.
Table 4 presents the estimates of Model 3. Our results from RE and FE estimates are similar.We fi nd that the average trade openness and fi nancial account balance with volatilities of domestic credit and investment have a signifi cant eff ect on the growth volatility.It seems that trade openness helps lowering the output growth volatility.We should note that an increase in the fi nancial account balance leads to a rise in the volatility.DGMM and SGMM estimates indicate that the volatilities of government consumption, trade openness and investment are signifi cant whereas, unlike static estimation results, the trade openness, fi nancial account balance and volatility of domestic credit are not.We also fi nd that the government consumption has a negative eff ect.We should note that the volatility of investment is the only variable that has a signifi cant eff ect in both static and dynamic estimations.Interestingly, the average and volatility of infl ation rate do not have a signifi cant eff ect any longer in both static and dynamic panel models when we control the volatility of our explanatory variables.Our results show the importance of employing both average values and volatilities of related variables in examining the output growth volatility.
Finally, in almost all GMM specifi cations, we fi nd that the lagged values of dependent variables are statistically signifi cant.Sargan test results suggest that we fail to reject the null hypothesis that is the overidentifying restrictions are valid.AR (2) test results do not indicate any autocorrelation problems.The number of instruments in our estimates is equal or smaller than the group numbers.We would like to point out that, although it increases the number of instruments, adding time dummies do not aff ect our results in general.We tend to consider our GMM estimates more robust and reliable compared to linear panel estimates, as explained in detail above.

Conclusion
Our study examines the relationship between the government expenditure and output growth volatility in 57 low and middle income countries by using static and dynamic panel estimation methods.Based on our dynamic panel estimates, which are more robust and reliable, we conclude that the average and volatility of government consumption and volatilities of trade openness and investment are signifi cant in explaining the volatility in low and middle income countries.These signifi cant variables have the correct or expected signs.More briefl y, average government consumption has a negative impact on the growth volatility while the volatilities have the positive eff ect.
We should note that our empirical results clearly highlight four important points.First, there exists a strong negative relation between the government expenditure and volatility for low and middle income countries.This fi nding strikingly diff ers from that of some studies in the literature, which do not report any signifi cant stabilizing impact of government expenditures on the volatility in developing countries.Second, all estimates indicate that the volatility of investment has a robust eff ect on the growth volatility.Third, it is important to include both averages and volatilities of related variables in the same specifi cation in examining the volatility.Fourth, our empirical results also show the importance of using dynamic panel methods in this context.
To have a more stable economy, our results suggest that policy makers in low and middle income countries should pay more att ention to some policies.First, in our opinion, it would be a good policy to make automatic stabilizers work bett er by changing the tax and expenditure systems while keeping the government size at the optimal level.Second, these countries should have a sound and prudent fi scal and monetary position, like low budget defi cits and infl ation rate, to eff ectively carry out the discretionary countercyclical policies when needed.Third, implementing some well-structured, clear and business friendly economic policies, based on a long term perspective and strong economic rationale, should be a priority to prevent the occurring of sudden and large fl uctuations in the expectations of investors and economic agents.Moreover, adopting a fl exible rule-based monetary and fi scal policy might be helpful to reduce volatility and eventually support growth by establishing credibility and providing confi dence.Such a policy would be especially more successful at keeping the government expenditures and investment more stable.Fourth, the eff ective protection of property rights and maintaining the rule of law would be decisive in smoothing particularly investment fl uctuations.In a broader sense, a vast literature convincingly shows the role or importance of institutions or institutional structure for the success of economic policies.There is no doubt that institutions matt er for economic outcomes, ranging from, for example, fi scal or monetary policy to economic growth.Therefore, improving institutional structure, public administration and policy making capacity should be an ultimate aim even though it is not an easy task.Otherwise, if institutions or institutional weakness are the main reason for economic

Table 1 :
Data description VolTOStandard deviation of trade openness (the sum of exports and imports as a percent of GDP) over non-overlapping fi ve-year periods AvCR Average of domestic credit provided by the fi nancial sector (as a percent of GDP) over non-overlapping fi ve-year periods AvGE Average of government consumption expenditures, as a percent of GDP, over non-overlapping fi ve-year periods AvGR Average growth rate of GDP over non-overlapping fi ve-year periods AvFA Average of fi nancial account balance (as a percent of GDP) non-overlapping fi ve-year periods AvIN Average of infl ation rate over non-overlapping fi ve-year periods AvIV Average of investment rate (Gross Fixed Capital Formation, as a percent of GDP) over non-overlapping fi ve-year periods AvTO Average of trade openness (the sum of export and import, as a percent of GDP) over non-overlapping fi ve-year periods