English / ქართული / русский /

Journal number 3 ∘ Lasha Arevadze
Estimation of Government Spending Multiplier in Countries in Transition


This paper examines theoretical and empirical approaches for estimating government spending multiplier for Countries in Transition, among them the focus here is Post-Soviet Countries (CIS countries). Government spending multiplier is low in developing countries, mainly it is caused with high debt level and weak institutions of those economies, CIS countries are not exception of these features. The paper studies the effects of government spending on CIS economies and how this effect depends on economic conditions, such as public debt level, cyclical position of economy, currency regime and etc. 

 Keywords: government spending multiplier, CIS economies 

Acknowledgment: I would like to show my warm thank to Professor Iuri Ananiashvili, who is my supervisor at the department of econometrics at Tbilisi State University. Also, I wish to present my special thanks to Professor Temirlan Moldogaziev and Professor James Monogan from University of Georgia (USA) for their comments and beneficial revision of the article. 

1. Introduction

The main purpose of this paper is to investigate evidence about fiscal policy (spending side) efficiency in CIS countries (Common Wealth of Independent States: Armenia, Azerbaijan, Belarus, Georgia, Kazakhstan, Kyrgyz Republic, Moldova, Russia, Tajikistan, Turkmenistan, Ukraine, and Uzbekistan). Limited numbers of publications are available for developing countries on fiscal multipliers. Actually, the shortfall of relevant evidences about developing countries stems from the data limitation. Moreover, time series on main macroeconomic variables are nonstationary and serial correlation is a serious problem.

The paper studies the effects of government spending on CIS economies and how this effect depends on economic conditions, such as public debt level, crisis time and exchange rate regime. In order to evaluate effects of spending shock to economy two stage model is applied in the paper. On the first stage, exogenous spending shock is identified and on the next stage its effect to economy is assessed based on fixed effect panel data model. Spending multiplier is not constant and it varies across policy environment, spending effect is stronger in crisis time, however, there are opposite cases as well. Also, bad fiscal time measured in GDP/debt ratio influence negatively on multiplier effect, as well as in case of currency peg regime the spending effect is stronger. The paper contributes to literature on fiscal policy in CIS economies by identifying government spending effect across different policy environment. 

After global financial crisis in 2008-2009 and the boost of Bubble on the assets market, we face sequential boost in the literature as general and about fiscal policy efficiency in particular. Interest among economists is due to unprecedented huge fiscal stimulation packages proposed by national governments for recovering their economies. In principle, the total amount of anti-crisis package was $2.18 trillion or 3.5% of world`s GDP. Existed literature about developed countries shows high efficiency of fiscal policy, i.e. high multipliers. In contrast, despite this boost in literature about developed countries, the developing countries still remain as a periphery of researches about the government spending multiplier and CIS countries are not exception.

There is no “the multiplier”[1] and the effect of government spending varies with economic conditions. The size of the multiplier is affected by crowding out effect of private consumption and investment, if this effect dominates, then multiplier is small and sometimes it can be negative. However, the effect varies in response to the economic situation. During economic and financial crisis fiscal policy is more efficient than in expansion, as there is  scarce liquidity during recession while government spending shock address this constraint. As well as literature shows spending multiplier is high in case of fixed exchange regime. Government spending shock results in higher interest rate, if financial market is open then it causes appreciation of local currency and export is declined in case of floating exchange regime, however, in case of fixed exchange regime government purchases foreign currency to defeat local currency from appreciation and there is no trade off in terms of export.

One of the stylized facts on multiplier is that it is smaller in developing countries. This means more crowding out effect of government spending in less developed countries. Also, in case of advance economies there is no or limited difference between government consumption and investment multipliers, but the multiplier effect of government investment is higher in developing countries. Another stylized fact is that in case of high public debt level, fiscal policy can be contractionary for economy and vice versa fiscal consolidation can be expansionary as consolidation results in lower risk premium of a country because of the lower debt burden. To sum up, the size of government spending multiplier is not unique number for a particular economy; moreover, its size is affected by various economic conditions.

Post-soviet countries have experienced one of the turbulent economic environment after the collapse of the Soviet Union, on the way from socialism to market economy the size of government declined substantially, they have experienced financial and currency crisis in 1998 and 2008, as well as there was wave of currency devaluation from 2014 in CIS countries, what is effect of government spending in those countries and how it contributes to GDP under such a turbulent economic environment is a main research question of this paper.

In literature various econometric models are applied for identifying government spending multiplier, among them VAR model (such as Ilzetzki and others (2009)) and Dynamic Stochastic General Equilibrium model (DSGE), (such as Blanchard and other 2002) are widely used. To evaluate government spending multipliers which vary across economic conditions such as crisis time, currency exchange regime, debt level and etc. In the paper two stage linear estimation technique is applied to identify government spending shock at the first stage and to evaluate conditional multiplier on the second stage.

The remainder of this paper is organized as follows: Section 2 reviews existed literature on identification issue of spending multiplier especially in developing countries. Section 3 briefly describes theory and methodology for estimating government spending multiplier, Section 4 presents data, Section 5   describes results from two stage fixed effect model for 12 CIS countries; Section 6 concludes. 

2. Literature Review

Limited numbers of publications are available for developing countries on fiscal multipliers. Actually, the shortfall of relevant evidences about developing countries stems from the data limitation in those groups of countries. Moreover, time series on main macroeconomic variables are nonstationary and serial correlation is a serious problem.

Despite those limitations, there still exists some analysis for developing countries and various econometric models are applied. Researchers have to work out a new manner to isolate fiscal policy shock in developing countries. This shock should be unanticipated and not related to the current economic situation, otherwise private agents, who internalize government budget constraint in their intertemporal constraint, change their behavior and estimation will be biased. One of the contemporary studies for the sample of 102 developing countries on estimation of government spending multiplier was applied by Kraay in 2012. Best of my knowledge this is largest panel of countries such analysis is applied for, because of data limitation, Kraay (2012) in his paper use instrumental variable procedure for determining fiscal shock. In principle, he uses a debt reimbursement as an instrument of government spending shock. Kraay tray to address problems related to policy anticipation and independence from the current economic situation by the assumption that debt reimbursement is determined in advance and it is not related to the current economic situation. Moreover, as Kraay shows, only 20% of the total loan is disbursed in the first period and 17% in the second year and so on. Resulted from this evidence, Kraay claims that disbursement of particular loan is predetermined, but as in developing countries loan is one of the basic sources to finance expenditures, then it can be a good instrument of government spending. In this group of countries, Kraay applies his analysis, on average 16% of expenditure is financed at the expense of loan taken from multilateral and bilateral official creditors. But this strategy cannot be taken without careful judgment. Firstly, Kraay has conducted this analysis about 102 developing countries, but approximately 84% of those countries use foreign debt as a significant source for financing government expenditures.  Hence, if we follow Corsetti`s analysis about fiscal multiplier, most part of developing countries are below -6% critical threshold of negative budget balance (i.e. net borrowing).

The multiplier identified by Kraay varies from 0.38 to 0.42; which is fairly lower then multiplier for developed countries. However, there is a question about the multiplier calculated by Kraay; is it really unconditional multiplier i.e. multiplier in normal times? If we take into account the fact that most of the countries, Kraay applies his analysis, use loans to finance its expenditure, it is a conditional multiplier in case of high indebtedness. As other studies show fiscal policy is less efficient in case of higher debt, and then we can claim that the multiplier identified by Kraay in developing countries makes difficulties to interpret as multiplier in normal times in those countries. Moreover, the size of multiplier is affected not only by debt level but also, the source of debt financing matters. The rest part of this section deeply review some stylized facts in the literature on government spending multiplier.  

Effect of debt level and its composition on the size of multiplier  

The indebtedness level of a country effects on the size of the fiscal multiplier and it is higher for countries with low levels of debt (Corssetti, 2012). Initially, when the public debt level is not very high government spending and private consumption are positively correlated, but when the debt level reaches a critical level they move in the opposite directions. The motivation of this argument is that a high level of debt is accompanied by a high interest rate because of higher risk premiums, especially in developing countries. Consequently, the crowding out effect dominates and fiscal policy efficiency is small. Furthermore, if the Debt/GDP ratio is above 50%, then fiscal policy is strongly unproductive; the long run multiplier in this case is -2 (Ilzetski, 2009). This once again means that to finance government consumption at the expense of taking debt is not a good deal. It is obvious that we face strong crowding out effect, which can be result of expectation about future consolidation and higher taxes, i.e Ricardian equivalence holds.

There exists some evidence about the strong crowding out effect of private investment due to public expenditures financed at the expense of domestic loans from banks. For instance, Emran and Farazi (2008) showed that if governments in developing countries take loans from  banking sector, then each $1 (taken from a bank) caused a reduction of private investment by $0.8. This evidence emphasizes the liquidity constraint in developing countries. Moreover, Gupta et al (2005) argues that a 1 percentage point increase of expenditures financed through domestic loans causes a reduction of the potential growth rate by 0.75 percentage points. The argument is that if deficit is financed through domestic debt, then it amplifies inflation, which negatively effects on growth.

It should be mentioned that the source of financing matters for the size of multiplier. In particular, debt financed expenditure causes an increase in private investment in developing countries; if such expenditure is financed through tax then it has a negative effect (Miller and Ahmed, 1999).  Miller and Ahmed (1999) analyzed the effect of source of spending on crowding out effect on investment and show that debt-financed expenditure crowds out investment if it concentrates on social security and welfare, but expenditures on communication and infrastructure crowds in. As for tax-financed expenditure, we face a negative effect on investment if it is spent on infrastructure.

 This idea is also supported by the analysis of Barro (1990), He states that tax-financed government expenditure causes a negative effect on GDP. Moreover, if it is productive spending, then it balances negative tax effect and causes a positive impact in the economy.  While argument in favor of debt financed expenditure shows a liquidity constraint in developing countries. A government provides additional liquidity to economy by taking loans instead of tax based expansion. However, we should not forget the discussion about critical level of debt and its negative impact on the effectiveness of public expenditure.

Composition of Government Spending

To determine the size of the fiscal multiplier is equivalent to identify when it crowds in or crowds out private consumption and investment. Existing literature about the effect of fiscal policy on investment can be summarized as follows:  government capital expenditure in developing countries crowds in investment, as private and public investment can be considered as complements, while government investment is not more effective than consumption in developed countries. The last finding fits the traditional view of crowding out effect of government spending due to higher interest rates. However, evidence about developing countries is a non-traditional view, which is more powerful under non full employment condition. However, the magnitude of this effect depends on the structure of spending. For example, Sadeghi et al (2013) in their panel analysis about developing and developed countries show that the elasticity of private investment with respect to public investment is positive and significant (0.31), while it is 0.18 in developed ones. Moreover, private investment elasticity w.r.t. other type of expenditures is negative and lower than in developed countries, i.e. crowding out effect. Those authors conclude that in developing countries public investment is more efficient because private agents face a liquidity constraint, and interest rate differentials. However, there exist some other evidences which do not support this idea. For example Hasan (1960) states that in developing countries public investment has a long run nature (infrastructural projects, irrigation and other) and it has a low effect on current output and it contributes to inflation. The paper by Hassan (1960) is only a theoretical discussion, but recent empirical evidence from the IMF suggests that public spending on investment has positive impact on the economy, not only in the short run, but also in the medium-run; Estevao and Samarke (2013) in this paper estimate investment multiplier as from 0.2 to 0.7 and from 0.42 to 0.92 in the short-run and in the medium-run, respectively, in case of central American developing countries, while non-productive expenditure has a negative impact on those countries. As for the type of expenditure, the empirical evidence about developing countries shows that cutting expenditures on investment can be more harmful in the case of economic downturn relative to cuts in current expenditure.

Devarajan et al (1996) showed that in the case of developing countries, current expenditures positively effect on GDP, while capital expenditures have a negative influence for the panel of 43 developing countries over 20 years. In their model, they also account for the total level of government spending; hence, in their model a one dollar increase of capital spending automatically means a dollar reduction in current expenditures. But in other models which do not account for total level of expenditures, then as Devarajan et al (1996) claim, the strong positive effect of capital expenditure can be a reflection of level effect of expenditure and not the relative advantage of capital over current expenditure. Actually, Miller and Ahmed (1999) do not control the total level of expenditure in their linear fixed effect estimation for 23 developing and 16 developed countries. Their result is more informative and different than the previous evidences. In principle, they showed that if expenditure on transportation and communication is financed through either taking loans or tax, then it has a positive effect on investment for developing countries, but we have zero or a negative impact on investment in developed countries.  However, if the government spends money on social security and welfare, it crowds out private investment. The work by Miller and Ahmed (1999), however, does not say anything about the gross effect on GDP due to different types of budget outlays.

In case of developing countries full effect of government investment is not reflected in multiplier. Kraay et al (2013) claim that higher  share of expenditure on investment does not mean that it is automatically resulted in the higher asset accumulation in developing countries due to inefficiency and waste. This analysis is supported by Kneffer and Kanck (2007). Moreover, Gupta et al (2011) numerically showed that a significant part of public expenditures on investment are not reflected in productive investment in developing countries. For example, in case of low income countries the share of capital stock adjusted with efficiency criteria is 30.1% of GDP, while the raw share is 71%. This finding magnifies our intuition that due to institutional rigidities, the spending on productive assets is not effectively used in those countries. Furthermore, the income share of adjusted public investment in GDP is lower (0.14 instead of 0.25), but because of downward adjustment of public investment by efficiency criteria, then marginal productivity for low income countries reaches 0.88, which is too high. Moreover, it is slightly higher than the marginal productivity of private investment (0.71).

To sum up, as it was previously stated in the analysis by Ilzetzki  et al (2010), the government investment multiplier is 0.6, which is significantly much higher than consumption multiplier (-0.19) in developing countries ; the paper clarifies that with 95% confidence level we can say that the efficiency of government investment is higher than the consumption multiplier in those countries. Moreover, the multiplier effect is further eroded with inefficiency.

Fixed Exchange Regime versus Flexible

In his paper, Müller (2012) identifies government spending multiplier as 0.75 under a floating regime vs. 1.2 under a pegged regime. The difference is significant but not as large as  in other works. For example, Petroviḉ et al (2014), for a panel of 10 emerging European economies show that the spending multiplier under a pegged regime is 1.31 versus 0.03 under fixed exchange regime. Moreover, Muller finds evidence against the traditional Mundel-Flaming channel; in this standard framework, government spending under the flexible exchange regime causes appreciation of the domestic currency and additional crowding out of net export, but in this evidence we face no deterioration of net-export. This means that there are other channels for making the difference between fiscal policy efficiency under fixed and flexible regimes.

There is no strong convention among economists about fiscal policy efficiency under fixed and flexible regimes. One cohort of economists for example, Corsetti et al (2012) claim that increasing outlays from the budget causes an increase of the interest rate. As a result, it causes currency appreciation, but if then monetary authority desires to fix the exchange rate, then it should increase money supply, which is additional stimuli for an economy and because of this we would expect a higher multiplier. But as other studies show, the real world is not so simple and we should investigate the effect of fiscal expansion under a fixed exchange regime and its interaction with other conditions of an economic environment. For example, Bonam and Lukkezen (2013) show that when a country experiences high sovereign debt risk, then currency devaluation is expected to the response of fiscal expansion. As a result, the monetary authority will have to tighten monetary policy in the case of a fixed exchange regime. Consequently, we face a higher interest rate and stronger crowding out effect of investment and consumption, hence a lower multiplier under a fixed exchange regime instead of the big stimulus predicted by one group of authors. However, Bonam and Lukkezen (2013) argue that there is a room for fiscal policy when its effect is higher under a pegged than flexible regime, even if we introduce sovereign risk in our analysis. For example, expansionary fiscal contraction holds only in the case of a fixed exchange regime only in the short run. Hence, if sovereign risk is high and fiscal authority consolidates, then it reduces the risk premium and stimulates an inflow of capital, which causes appreciation of the domestic currency. Under the peg, however, monetary authority intervenes and we do not face crowding out effect of export. This last finding once again emphasizes the importance to account existed economic environment during discussions on spending multipliers.

The difference between multipliers under fixed or flexible regime is also determined by degree of capital mobility. Riguzzi at el (2015) find that the output multiplier is lower (0.71) if capital markets are well functioned, relative to the case when capital mobility is limited (1.44). The intuition is   that when capital mobility is limited then to the response of fiscal expansion exchange rate appreciates, but if capital mobility is limited, then the degree of possible appreciation is lower and as a result crowding out effect of export is minor, hence multiplier is higher. If we account imperfect capital market in the developing countries then this factor would have a positive effect on the size of spending multiplier. Moreover, Kraay (2012) show that in case of developing countries there is no difference between multipliers under fixed  or flexible regime, because if financial market aren`t well functioned then there is no appreciation of domestic currency and no necessity of intervention. Hence, financial openness negatively effects on spending multiplier as well trade openness causes the lower multiplier. 

Effect of openness on the size of multiplier

There is abundant empirical evidence about the relationship between openness and the effectiveness of fiscal policy, for example,  Corsetti et al. (2012), Ilzetzki et al (2011) and etc. Also, Karras (2011) conducted analysis for 62 developed and developing countries and show that fiscal multiplier is decreasing function of degree of openness. For example, when openness is around 10%, then multiplier is too high (1.39) for this group of countries, but if openness is 50%, then multiplier becomes 1.05 and in the very extreme case when openness is more than 100% then multiplier reaches the lowest level (0.61). If the degree of openness is high, then the significant part of purchases by government are concentrated on imported commodities and the impact of policy on the domestic economy is limited. Above mentioned findings can be summarized with the table I.

To sum up we face significant variation of fiscal multipliers across various policy environment, the above discussion supports Corsetti`s idea that there is no “the multiplier”. Consequently, in this paper we analyses government spending multiplier by accounting policy environment, as it is suggested by Corsetti and others (2012).  

Government spending multiplier in developing countries

There exists a significant evidence gap in the literature about the size of the spending multipliers in developing countries, while it is widely researched for advanced economies. Nevertheless, some evidence about fiscal policy efficiency in developing countries still exists. For instance, the short run multiplier for developing countries is 0.4, (Kraay (2012)); while it is higher and close to one in the developed countries (Corssetti, 2012). As a result (see, table I

I), first  stylized fact from literature about developed and developing countries is that fiscal policy efficiency is higher in developed countries.

The basic reason why empirical evidence about developing countries is scarce is in lack of precise and long time series data for government spending as well as for other relevant variables to estimate fiscal policy efficiency. Due to this, for example, Kraay (2012) uses different methodology to estimate the multiplier in developing countries. In principle, Kraay in his paper uses an external debt reimbursement as an instrumental variable of government spending, because this variable is well accounted by the World Bank for each country and it is strongly correlated with government spending in developing countries. 

Table I. Summary of literature on fiscal multipliers.



Impact multiplier




Developed countries[2]

Perotti (2005)





Government spending













Romer and Romer (2008)


1.2 (one year)

4 (cumulative)



Ramey (2008)





Gov. spending

Johnson, Souleles, and Parker (2006)





Tax rebates

Ilzetzki and Vegh (2008)

High income




Government spending

Blanchard and Perotti (2002)





Government spending

Cogan and others (2009)



1.2- 1.5


Government spending

Dalsgraard, Andre, and Richardson (2001)


1.1 -1.5(one year)



Government spending


1.7-2.6 (one year)


Euro Area

1.2-1.9 (one year)


Elmendorf and Furman (2006)





Government spending

Corsetti and others (2012)




Two step model

Government spending


Developing and emerging economies

Kraay (2012)

102 developing countries



IV estimation

Government spending

Ilzetzki  and others(2009)





Government spending

Ilzetzki (2011)





Government spending

Shen and S.Yang (2012)

Developing countries




Government[3] spending




Government[4] spending

Petrovic and others (2014)

Emerging European economies





Government spending

 3. Theory

In this section, I review different theoretical approaches to identify fiscal multipliers; i.e. what they are and why they vary across different theory. Ilzetski and others (2012) claim that there is no “the multiplier” and it varies across different policy environment. Therefore, I will review key drivers of the size of multipliers resulted from the different economic environment and finally based on theoretical discussion; I will formulate two stage procedures for estimating government spending multipliers, similar to the model developed by Ilzetski and others (2012).

The simplest definition of a fiscal multiplier is that “it [spending multiplier] is the change in output due to a change in fiscal policy instrument” (as cited in Chinn 2013). Algebraically, it is ∂y/∂x where y is output and x is a policy instrument. If we wonder about the immediate impact of a policy, we should calculate the impact multiplier (i.e. ∆Y/∆G), but in most cases, especially when we try to calculate fiscal multiplier based on quarter data, we should calculate a cumulative multiplier:

Most of empirical evidences about multipliers use not only Yt (GDP) as an endogenous variable, but also, researchers try to investigate how a change in fiscal policy is reflected in changes in consumption, investment, current account balance, exchange rate and other macroeconomic variables (Corssetti, 2012; Petroviḉ, 2014).

I begin the discussion about the fiscal multiplier from the perspective of Neoclassical Synthesis. This theoretical model is characterized by Keynesian properties in the short run and Classical properties in the long run. As a result, as prices are sticky and cannot adjust quickly in the short run, fiscal policy is efficient and the multiplier is larger. Nevertheless, we expect no permanent effect from the change of fiscal policy as we use properties of Classical theory in the long run (Chinn, 2013).

The government spending multiplier is close to zero in Neoclassicists` approaches, as they assume no nominal rigidities. Moreover, this theory predicts a negative fiscal multiplier if taxes are distortionary.

Finally, as a New Keynesian model uses inter temporal optimization, it is close to Classical approaches in its predictions, but additionally the model uses the assumptions about nominal and real rigidities. Due to these last extensions, fiscal policy has an effect on output in the short run (Chinn, 2013). Fiscal policy is not secured itself from a negative spending multiplier if expenditure is expected to be financed at the expense of taxes in the future; this is reflection of Ricardian equivalence (Capet, 2004).  An expectation that taxes will be higher in the future erodes consumption and investment today and output decreases consequently, so there is a negative spending multiplier.

In order to estimate multiplier in practice at least two conditions must be satisfied for government spending shocks, it should be unanticipated by economic agents and contemporaneous exogenous. Moreover, there is no “the multiplier” and it varies in relation to the economic environment. First of all cyclical nature of government spending matters.  If government spending is acyclical, then the requirement of contemporaneous exogeneity holds. But we still do not have a guarantee that fiscal policy is unanticipated.  Empirical literature about the cyclical nature of fiscal policy can be summarized as follows: fiscal policy is acyclical or counter cyclical in developed countries and mostly pro cyclical in developing ones (Ilzetzki, 2008). For example, the measure of the cyclical nature of government spending is 0.61 for developing countries and -0.11 for developed ones based on GMM estimation. Kraay et al. (2013) make a distinction between discretionary fiscal policy and automatic stabilizers and claim that in the case of developing countries, both of them are pro cyclical; automatic stabilizers are ineffective in those countries to smooth fluctuation of output. Moreover, they state that the procyclical fiscal policy is a reflection of constraint on financial resources, and also, sometimes it is politically motivated. Kraay points out that during the post crisis in 2008, the developing world and their governments had a better fiscal stance and they had a tendency to  more counter cyclical policy then in the pre-crisis period. Theoretically, counter cyclical policy is optimal, as increase in government spending relaxes liquidity constraint in economy in case of recession and its effect is stronger. Hence, recommendation is that government should increase spending in crisis time and reduce it in boom. In contrast to last argument about the advantages of counter cyclical fiscal policy, Hemming (2002) advocates that contractionary fiscal policy can be expansionary. Hence, fiscal consolidation causes a positive response from the private sector. The basic reason is that consolidation, in particular, if a country faces high public debt, causes reduction of risk premium, and as a result, the crowding out effect will be lower after consolidation and fiscal policy recreates power. Moreover, if low and negative fiscal multiplier is tendency in developing countries, then pro-cyclical fiscal policy seems rational for those economies; moreover, if they have problems related to solvency. For example, if a particular country increases its expenditure in bad times and the multiplier is negative, then it causes an additional reduction in GDP due to counter cyclical fiscal policy. Based on this discussion pro cyclical fiscal policy takes advantages in the case of developing countries mainly during recession. Debt level is main reason why counter cycle fiscal policy is not optimal in case of developing countries, high debt level bears positive and high risk premium from investor’s point of view, hence, if public debt is high then better strategy is to reduce government expenditure. In addition to low multiplier in case of high level of public debt, if it is financed through domestic loan then the multiplier is lower than if it is financed through foreign loan. Intuition is that if financial resource is scarce within country then additional financial resource from  abroad provides additional liquidity to economy and crowing in effect of government spending is larger.

Not only debt composition matters in terms of size of multiplier but also spending composition is important, in developing countries public  investment is more efficient then government consumption, as public investment has complimentary effect and crowds in private investment, as a result effect is stronger.

In case of fixed exchange rate multiplier is higher, as expansionary fiscal policy provokes expansionary monetary policy. However, situation changes if capital market mobility is low. Higher government spending financed with  debt causes higher interest rate that can be reflected in inflow of financial resources from abroad, it pushes exchange rate up, and monetary authority has to increase supply of money to reduce interest rate differential. But if capital mobility is low then there is no pressure on    exchange rate and no necessity for monetary expansion as a result no additional multiplier effect. 

Based on empirical strategy described below size of spending multiplier will be iterated with different macroeconomic conditions. 

Empirical strategy

The existing literature on government spending (or consumption) multipliers are mostly concentrated on identification issue of discretionary fiscal policy shock, i.e. the unanticipated part of government expenditure and its effect on output, private consumption, exports and many other macroeconomic variables. To identify the unanticipated part of government outlays is a principal requirement for precise estimation, especially for the research concentrated government spending multiplier based on VAR (vector autoregressive) model. In order to identify the unexpected part of government consumption SVAR (stochastic specification of VAR model), models make some linear assumptions about fiscal policy rules, and the rest of government spending can be considered as the unanticipated part. But as Ilzetki (2010) points out, there is no guarantee that the unexplained part by econometric models cannot be predicted by private agents; this is common across other papers based on the conventional VAR estimation developed by Blanchard and Perotti in 2002. But if the fiscal shock is predicted by private agents, this causes changes in their behavior and we cannot estimate the efficiency of fiscal policy precisely.

There are also a few papers, for example, Ramey (2011) which use extraordinary shocks in the economy which cannot be anticipated. For example, Ramey uses war time expenditure  in the case of the USA to predict the  government spending multiplier. Ramey argues that military expenditures in the case of the USA were properly exogenous shock. Other findings notify us about that the  exogeneity assumption about war time requisition is ambiguous. For example, Ilzutski (2008) in his paper argues that war time can be predicted in advance by individuals. The reason why it is problematic is that when we want to predict exogenous fiscal shock there are two basic requirements. Firstly, as we already said, fiscal shock should be unanticipated and secondly, shock should not be related to the current economic situation. Without ambiguity it can be said that wartime requisition is absent from the current economic situation. But it can be anticipated by private agents, because war cannot be started immediately. Moreover, the main reason why the model proposed by Ramey contains potential problems is that it requires a long time series, For example, the analysis by Ramey is applied over a long period of history of the USA and it covers a war time in Vietnam as well  in Korea. Such long data is not  available for most of other countries; moreover, those wars were out of the territory of the USA and these war times were not related to a significant reduction of capital in this particular country. Of course, this is not the case in other countries where the result of war was an extraordinary destruction of capital. Those are the main drawbacks of the model proposed by Ramey; at least it is too problematic to apply this model for other countries besides the USA.

As we say, the basic requirement is that fiscal shock should be contemporaneously exogenous, hence, fiscal stimulus should not be driven by the current economic situation. Because of this,  researchers have to make an  assumption about the lag in which fiscal policy can react to contemporaneous change in the economy. In normal times, of course, this lag is just one year; as the fiscal authority works out a new budget plan within one year intervals; but more problematic is that what happens during sudden and sharp downturn or boom periods. Most researchers who use quarter data claim that one quarter  lag is quite relevant in this case; i.e. fiscal authority needs at least one quarter to adjust its initial budget plan. The assumption that fiscal shock is contemporaneously exogenous is difficult to apply in the analysis which are based on annual data; in this context contemporaneously exogeneity means that the fiscal authority does not make any adjustment in the budgetary plan during one year. As Corsetti claims, a one year period as an implementation lag is relevant if there is no extraordinary case; for example, as this was in the case in the global financial crisis in 2008.

The second part of literature for identifying fiscal multiplier is based on New-Keynesian DSGE models. Structural models are not sensitive to assumptions required by VAR model, but they still bear some restrictions. The main assumption is related to price setting behavior. DSGE framework follows Calvo type price setting; according to this, one part of prices on intermediate goods are rigid in the  short time period, but some part of prices are free to optimize to the response of change in the economic situation. For example, Christiano, Eichenbaum, Rebelo (2009) in their paper for identifying the spending multiplier makes the assumption that 0.85 parts of price are rigid. And only under such exaggerated assumption they make a conclusion that the fiscal multiplier is 1.05. It is obvious that the size of the fiscal multiplier identified by DSGE models is sensitive with respect to the assumption about Calvo type price setting; if prices are less rigid then DSGE identifies a smaller multiplier against higher rigidity when the identified multiplier is close to Keynsian multiplier, and it is not a big surprise in the case of high price rigidity.

Moreover, there is no convention among economists about fiscal policy rule, which is one of the basic equation in DSGE framework  (Batini et al, 2014);  in principle, we observe different modelling of fiscal rule in different papers; this can be a potential source of variation of the fiscal multiplier itself.

Those problems related to conventional VAR and DSGE models is amplified by data limitation in the case of developing countries, this is a main reason why we have little empirical evidence for this group of countries. As we face lack of convention in estimation procedures and huge variation among calibrated multipliers is tendency, Batini et al (2014) provide a new model for estimating multiplier under data limitation called “the bucket approach”. This model is basically an “analytic guess” about the fiscal multiplier. The multiplier under this model is a simple weighted average of scores assigned to particular state of country (openness, rigidity of the labour market, level of automatic stabilizers, fixed or flexible exchange regime, level of government debt, effective expenditure/revenue management), but the authors allocate this measure of multiplier to three different intervals, hence, it gives us interval estimation and not point estimation of the fiscal multiplier. Unfortunately, except its simplicity under data limitation, this model has no theoretical background.

Finally, we can say that each type of model, i.e. VAR and DSGE models cannot be used without pros and cons, and we need deeper analysis of specific application of those models to determine when the identified multiplier by economists can be obtained without any significant bias. 

As it is stated in many papers, for example Perotti (2007), Muller (2012) the conventional VAR models proposed by Blanchard and Perotti (2002) are unable to predict fiscal shock, which is perfectly discretionary or unanticipated. Perotti`s argument is that most of the papers use one quarter implementation lag which is too short for fiscal policy to be implemented in this period, because of long implementation lag it can be anticipated by private agents. This idea is tolerated by Muller et al (2012). They proposed to add predicted government spending in model. Due to this, we can reduce the risk of anticipation. This approach is practically implemented by Corsetti (2012) for OECD countries when he uses composite leading indicator as a proxy of predicted GDP in his two stage regression. As long run time series  about predicted GDP or government spending is not available in the case of developing countries, then economists will have to work out new ways for predicting unanticipated fiscal shock, for example, Kraay use instrumental variable estimation  (but as was discussed, his approach is not free from problems).


Corsetti, Meier, and  J. Muller (2012) suggests  a two-stage model to identify fiscal multiplier. This model tries to solve omitted variable problem which is the main problem in conventional VAR models. In principle, they put dummy variables which describe the economic environment in the model, such as: currency peg dummy, crisis dummy, the state of the primary balance of budget, public debt level.

The objective on the first stage the model identifies fiscal shock which will be orthogonal to other developments in the economy, i.e. exogenous shock of government spending. At this stage this model assumes that the fiscal policy rule can be described with past information on the economic environment. In principle, the explanatory variables at this stage are: trend variable, two lagged values of government spending (gt-1, gt-2); two lagged value of real GDP per capita (Yt-1, Yt-2); public debt/GDP ratio (bt-1); peg dummy (pegt-1,i); the dummy variable which shows the level of the primary deficit (straint-1,1), crisis dummy (crisist-1,i). But as it pointed out in other papers fiscal shock which is generated through fiscal policy rule  can be anticipated if the rule is based on only past information, and public expenditure can be predictable in this case. Due to this Corsetti and others try to control future development in the economy by introducing composite leading indicator (clit-1,i) which is used as  a proxy of predicted GDP. The dependent variable is government spending per capita in t period.

Stage (1) 

  Gt,I ii *trendti,1 gt-1 + βi,2 gt-2i,1 Yt-1 i,2 Yt-2iclit-1,Ii bt-1+ ρi,1 pegt-1,1 + ρi,2straint-1,1 + ρi,3 crisist-1,1t,    (I) 


At the second stage, this model uses fiscal shock and its lagged values as explanatory  variables (μhatt,I, μhatt-1,I ,μhatt-2,I ,μhatt-3,i) generated from the first stage regression. Also, interaction between fiscal shock and dummies which describes economic environment is applied (μhatt,I*dt,I            μhatt-1,I*dt-1,I , μhatt-2,I*dt-2,I , μhatt-3,I*dt-3,I ), as well pure effect of dummy variables on the dependent variables is considered (dt,I,dt-1,I,dt,i-2,dt-3,I ). The dependent variable can be firstly, GDP per capita, as well as, consumption, investment, export, exchange rate and many other macroeconomic variables.

Stage (2)   

  Xt,i =ai +kitrendt +qixt-1,I iμhatt,I + ξ2μhatt-1,I + ξ3μhatt-2,I + ξ4μhatt-3,I1(μhatt,I*dt,I )+ τ2(μhatt-1,I*dt-1,I )+ τ3(μhatt-2,I*dt-2,I )+ τ4(μhatt-3,I*dt-3,I )+ω1dt,i+ ω2dt-1,i+ ω3dt,i-2+ ω4dt-3,I t,   (II)

Corsetti and others (2012) use this model to estimate the spending multiplier for OECD countries, based on annual data. In this section I am going to apply this model for CIS economies. The main limitation is the lack of sufficient data; I use annual data from 1991 to 2016 on GDP per capita (current USD), instead of dummy variable of bed fiscal time which is defined based on primary balance in the original paper, I use GDP/public debt threshold. Also, the data about composite leading indicator is not available for those groups of countries, and instead of this, I use a one year forward lag value of the trend component of GDP. 

4. Data

In this paper government spending multiplier is estimated with two stage linear model. On the first stage government spending shock is identified based on the fiscal policy rule as it was described in empirical strategy, fiscal policy rule is determined with public debt level, budget balance, currency peg regime, and crisis periods. Detailed description of those variables and their sources is given in the table II.


Table II. List of variables


Data source

Log of per capita GDP

World Bank database, World Development Indicators: GDP per capita (constant 2010 US$) World Development Indicators. Data is available from 1992 to 2015, in total we have 290 observations

Log of real per capita government spending

World bank database: World Development Indicators, general government final consumption expenditure (% of GDP) is used together with per capita GDP to reconstruct per capita government expenditure, data are available from 1992 to 2015, in total we have 288 observations.

General government gross debt (as % of GDP)

IMF, Data is available from 1992 to 2015 only for Russia and Ukraine, also, data on debt is available from 1995 to 2015 for most CIS countries. In total we have 256 observations for 12 countries

Financial crisis dummy

Takes on value of 1 during financial crises, and 0 otherwise, when one of them Systemic Banking Crisis (starting date) Currency Crisis,Sovereign Debt Crisis happens. The data comes from IMF Working Paper "Systemic Banking Crises Database: An Update"  by Luc Laeven and Fabian Valencia (2012) which is available from 1992 to 2012 ,            In total we have 262 observations for 12 countries.

Bad fiscal times dummy

Takes on a value of 1 when lagged public debt exceeds 100 percent of or alternatively time when General government net lending/borrowing (Percent of GDP) is less than -6%. Because of limited data availability on General government net lending/borrowing, we have to use public debt threshold for determining bed fiscal time. The 100% threshold is taken by Corsetti at al. (2012) for advanced economies, but 60% threshold is more prudent for CIS countries.

Peg dummy

Carmen M. Reinhart database: Exchange rate regime classification, annual, 1946-2016, (http://www.carmenreinhart.com/data/browse-by-country/ ) is used  as a peg dummy variable

Data is available from 1992 to 2015, in total we have 288 observations.


5.Results and discussion

On the first stage 12 individual regressions were run and residuals (spending shocks) were identified. Residuals from the first stage linear regressions are government spending shock, i.e. part of government spending that cannot be explained by fiscal policy rule.

On the second stage initially unconditional multiplier was estimated, hence, the model was estimated without putting dummy variables on economic condition in it. On the second stage firstly unit root test was applied for logarithmic value of GDP and as it was predicted non-stationarity was detected with Im–Pesaran–Shin test, but first difference of ln(GDP) is stationary. Hence, on the left hand side dependent variable in our model is growth rate of GDP. Unconditional fixed effect model predicts that coefficient of impact of government spending over growth rate of GDP is 0.045, however, this cannot be interpreted as multiplier, because the multiplier is level change in GDP from 1 dollar change in government spending; as the average share of government spending is 15% in CIS countries, we can say that 0.045/0.15=0.3 is unconditional government spending impact multiplier for CIS. Hence, one dollar government spending shock cause 0.3 dollar increase in GDP, this coefficient is quite close to multiplier identified by Kraay. First and second year lag multipliers are insignificant, on the third year impact multiplier is negative and significant, as a result cumulative multiplier is smaller than impact multiplier, this finding suggests that government spending is significant contemptuously, but close to neutral in the long run.

Table III. Second stage regression analysis



(model 1)


Debt level

(model 2)



Fixed vs. flexible exchange regime (model 3)


Recession vs. expansion

(model 4)











Spending shock









Spending shock -1









Spending shock -2









Spending shock-3









Spending shock*debt dummy









Spending shock1*debt dummy1









Spending shock2*debt dummy2









Spending shock3*debt dummy3









Debt dummy 









Debt dummy -1









Debt dummy -2









Debt dummy -3









Spending shock*peg dummy









Spending shock1*peg dummy1









Spending shock2*peg dummy2









Spending shock3*peg dummy3









Peg dummy









Peg dummy 1









Peg dummy 2









Peg dummy 3









Crisis dummy[7]









Crisis dummy 1









Crisis dummy 2









Crisis dummy 3









Spending shock* crisis dummy









Spending shock1* crisis dummy1









Spending shock2* crisis dummy2









Spending shock3* crisis dummy3



























 One of the core question in our model was how economic environment effects on the size of multiplier.

Effect of public debt level

60% of debt to GDP ratio was taken as a threshold to analysis effect of government spending when it keeps low level of debt versus higher level of debt. When the debt/GDP ratio is less than 60% (low level) then impact multiplier is 0.35[8]. Coefficient of interaction between government spending and debt dummy is not significant contemporaneously. Hence, impact multiplier at high debt versus low debt  is not different. However, coefficient of first and second year lags of interaction variable is significant, but signs of those coefficients oscillate. Hence the model doesn’t provide enough evidence to conclude that fiscal multiplier is smaller in case of high debt versus low debt level in the case of CIS countries, probably, it happens because of low frequency of sovereign debt problems among those countries in data.

Crisis and spending multiplier

I defined crisis as negative deviation from HP trend of GDP, hence the negative output gap, while positive deviation is expansion. Government spending impact multiplier in crises time is 0.13[9], this is significantly smaller than unconditional multiplier. Coefficient of interaction variable of government spending and crisis dummy on impact is not significant; hence, impact multiplier is not different across crises versus non crises conditions. However, the previous year multiplier in case of expansion  is positive and significant (0.0363), this means that if government increases expenditure in crises time then it will have negative effect on GDP in the next year (negative multiplier is 0.24). This evidence supports the idea on expansionary fiscal consolidation. Hence, if government will consolidates budget in crises time it has expansionary effect on GDP. Alternatively, I defined sever crisis as negative one standard deviation from HP filter. In this least case impact multiplier in non-crises time is 0.28 while coefficient of interaction variable is insignificant too.

Currency regime and the size of multiplier

Based on Carmen M. Reinhart database on currency regimes I have applied interaction between currency peg dummy variable and  government spending shock to assesses variation of multiplier in flexible versus fixed exchange rate regime. Impact multiplier in case of fixed exchange rate regime is 0.33, but model failed to identify different impact multiplier in case of flexible exchange rate regime. However, government spending shock has smaller effect in case of flexible regime versus fixed regime (multiplier is smaller by 0.23) after one year when fiscal shock is implemented, probably, this means that monetary expansion channel needs time to react on fiscal policy changes.

6. Conclusions or Implications

There is scarcity of researches on government spending efficiency, and developing countries are not exception. This paper analyses government spending multiplier in CIS countries. Main challenge for economists to measure government spending multiplier is to identify spending shock which will be unanticipated and contemporaneously exogenous. This paper use two stage linear model approach , on the first stage spending shock is identified based on modeling fiscal policy rule, and on the next stage spending shock is used as policy variable to estimate multiplier for the panel of 12 CIS countries.

Government spending unconditional multiplier identified in this paper is 0.3 for CIS countries. In principle, the multiplier is close to other estimates for developing countries. It shows that 1 dollar unexpected increase in government expenditure is resulted in increase of GDP by 0.3 dollar. As a result there is no evidence of expansionary effect of government spending in CIS countries.

Together with unconditional multiplier the paper also identifies spending multiplier across different economic condition. As model shows there is evidence that in crisis time government expansion can lead negative consequence to economy, it supports the idea on expansionary fiscal contraction. According to the model there is no evidence on variation of multiplier across debt level, while model estimates smaller multiplier in case of flexible exchange rate regime versus fixed regime. There is no strong evidence on lagged effect of government spending, hence, cumulative multiplier (long run multiplier) is not higher then impact multiplier which is close to 0.3 for the panel of CIS countries, as it was expected spending multiplier is low in this group of countries and this estimation is close to empirical evidences for other developing countries. To sum up based on the evidence of two stage models for CIS countries; fiscal authorities from those countries should not expect significant effect of government spending shock to their economies, in addition, there is evidence that fiscal consolidation contributes positive response of economy in crisis time in those group of countries.


Batini, Nicoletta, Luc Eyraud, and Anke Weber, (2014). “A Simple Method to Compute Fiscal Multipliers”, IMF working paper WP/14/93.

Chriastiano, Lawrence, Martin Eichenbaum, and Sergio Rebelo. 2009. “When is the Government Spending Multiplier Large?” National Bureau of Economic Research working paper 15394

Corsetti, Giancarlo, Andre Meier, and Gernot J. Müller. 2012. “What Determines Government Spending Multiplier?” IMF working paper WP/12/150

Estevao, Marcello, and Issouf Samarke, (2013). “The Economic Effect of Fiscal Consolidation with Debt Feedback”, IMF working paper WP/13/136.

Ilzetzki, Ethan , Mendoza Enrique G. and VÃegh, Carlos A. 2013. "How big (small?) are fiscal multipliers?," Journal of Monetary Economics,vol. 60(2), pages 239-254

Ilzetzki, Ethan, Carlos A. Vegh. 2008. “Procyclical Fiscal Policy in Developing Countries: Truth or Fiction?” National Bureau of Economic Research Working Paper 14191.

Kraay, Aart. 2012. “Government Spending Multiplier in Developing Countries.” The World Bank Policy Research Working Paper 609

Petroviḉ, Pavle, Milojko Arsiḉ, Aleksandara Nojkoviḉ. 2014. “Fiscal Multiplier in Emerging European Economies.” Fiscal Council of Republic of Serbia Research Paper.

Shen, W.  Shu-Chun S. Yang, and Luis-Felipe Zanna .2015.“Government Spending Effects in Low-Income Countries”, IMF working paper WP/15/286

[1] Corsetti at al. 2012. “What Determines Government Spending Multipliers?”, International Monetary Fund WP/12/150

[2] All those multipliers are taken from IMF staff paper note about fiscal multiplier (2009)

[3] Source of expenditure is domestic debt

[4] Source of expenditure is external debt

[5] Coefficient

[6] t-test

[7] 1 –in case of expansion and 0- in case of recession.

[8] Elasticity of GDP w.r.t. government spending is divided by share of government spending in GDP 0.0529/0.15

[9] 0.019/0.15