Although famous structural reforms are on the agenda of the growth debate, attention has returned to the benefits of counter-cyclical fiscal policy during a crisis. This post focuses on this issue, its mechanics and how it can best be pursued in the Euro-Zone. The conclusion is that any expansionary policy, if it is succeed cannot be pursued at the national level, but must be pursued at the Euro-zone, if not EU, level. Even then, certain caveats about the level of debt and the openness of the economy, among others should be considered carefully. However, a common fiscal policy would solve the inefficiencies highlighted by OCA theory, solve coordination problems and create a fiscal principal with the (democratic?) legitimacy to force monetary policy to accommodatingly “lean against the wind”.
This post is provided in the hope that it will make the large stock of information relating to this topic and its discourse somewhat clearer. In the following lines I begin by providing some of the intellectual, economic and political context for this debate. I move on to consider the theoretical models and the logic behind this issue. Then, I consider the empirical controversies around this topic. Finally, I conclude by summarising some of the results in the literature. In keeping with the geographical focus of this website, I make specific reference to estimates by Burriel et al (2010), who provide one of the most comprehensive studies for the Euro-Zone.
Although the overview is exhaustive, no claim is made as to its completeness. All links to the relevant sources are provided so that readers may skim through the sources themselves and take the relevant conclusions.
I leave the discussion of the econometric technicalities behind the estimation of multiplier for a later post, exclusively dedicated to that issue.
I – Context of the Debate
Since the advent, of the financial crisis an old debate has reappeared which had been all but relegated to the lesser pages of the history of economic research. As Perotti argues, while monetary policy, its transmission and effects dominated economic research during the great moderation and became well understood, a similar interest and resulting consensus does not exist around discretionary fiscal policy. Indeed, to this day although there is a general consensus that an unexpected fiscal policy shock will have positive effects on output, the field is still relatively divided as to the effects of that shock on consumption and the real wage. While theoretical models of a neoclassical penchant might predict a fall in consumption and real wages, supported by the empirical results of the natural experiments, new Keynesian models predicting an increase in consumption and real wages can be supported by the Vector Autoregressions.
This issue has become particularly relevant since December 2008 when the Federal Reserve’s decision to lower the federal funds rate and the discount rate virtually towards the zero lower bound, effectively made full use of the interest rate channel, thus threatening the economy with a liquidity crisis. The exhaustion of traditional policy tools created renewed interest in fiscal policy, its multipliers, their size and direction as possible tools to restart a stagnant economy. This interest was further rekindled by the Romer and Bernstein report, whose large multipliers sparked a controversy that led to a race to estimate multipliers and to a thriving academic and political debate.
This is not a frivolous discussion. Indeed, if the fiscal multiplier were as large as some of its supporters claim, then postponing its use is not only inefficient but morally reprehensible given the plight endured by the world economy since the fall of 2008. However, if its detractors are correct, expanding the size of the public sector could compound the largest economic crisis since the Great Depression. Therefore, understanding this policy tool, its scope and limitations, its transmission and ultimately its effect is more important now than ever before in the last 50 years.
This issue is even more relevant in the European Union, specifically for its members who have chosen to integrate further through the economic and monetary union (EMU), the Euro-zone. Whereas the USA already possesses the necessary tools to pursue such a policy, the question in Europe is not just to decide whether to use fiscal policy but whether this might be more efficiently used if it is pursued at an aggregate supranational level rather than at the traditional national level. Particularly, in light of the ongoing sovereign debt crisis, it is not trivial to ask, as De Grauwe does, whether a Eurobond would not benefit from the same liquidity advantages as US Treasury bonds and fluctuate less than national bonds do. If that were so, clearly such a further step in European integration would be warranted, as a way to mitigate debt effects on fiscal multipliers which would otherwise impose a lower limit on the scope for intervention. This would be a substantial contribution to the argument in favour of fiscal integration in the Eurozone. Until now, the support for this step forward in European integration has relied on considerations about mitigating asymmetric shocks in the context of the optimum currency area (OCA) debate, as discussed by De Grauwe among others, and increasing the size of multipliers, according to Forni and Pisanni. Therefore the answer to this question could have considerable repercussions for the Euro-zone.
II – Theoretical Foundations of the Fiscal Multiplier
In this section, I provide a survey of the literature. I describe the mechanics of fiscal multipliers within the model dynamics of real business cycle as well as new Keynesian models.
Theoretically, estimates of the effect of fiscal policy on economic growth are achieved through what are known as DGSE models. This approach is based on more sophisticated versions of New Keynesian macroeconomic models with which graduate students are familiar. They assume an economy with a number of individual, structural and environmental characteristics and find the economy’s equilibrium and deviations thereof based on utility and profit maximisation subject to budget constraints. These models are extremely advantageous because they are micro-founded, internally consistent and therefore traceable, while remaining extremely sophisticated. However, the necessary calibrations and the number of equations and parameters can make them both intractable and biased and have led to diverse and even contradicting estimates of fiscal multipliers. On the one side, stand studies such as the ones by Coenen, et al (2010), Romer and Bernstein (2009) and Zandi (2008) who find positive multipliers, with a size superior to 1 and with lasting effects. Alongside these studies stand the ones of Elmendorf and Furman (2008), Freedman, Laxton and Kumhof (2008), Heathcote (2005), HM Treasury (2003), Cogan et al (2010), Cwik and Wieland (2010) and Baxter and King (1993) which find a positive multiplier, but lower than 1. Finally Uhlig (2010) and Uhlig and Drautzburg (2011) find a negative multiplier. Although this multitude of estimates can create some confusion and suspicion, the differences can be easily understood.
Leeper, Traum and Walker (2011) attempt to shed some light on the “morass” of DSGE estimates through the use of Bayesian methods and prior analysis. This, they argue, following Geweke (2005), “is important for any macroeconomic question in which DSGE models are employed, whether or not Bayesian estimation is used”. The authors explore the effect of government spending on output, private consumption and investment. They find that the results arrived at through this theoretical approach are contingent on the relevant constraints imposed on the model, which bias the probability of finding positive multipliers. Using a nested model to study the effect of such restrictions, the authors find that RBC models, similar to the ones used by Uhlig (2010), whether benchmark or with frictions, tend to automatically impose a negative range for the fiscal multiplier. In these models, the Ricardian equivalence prevails and an unexpected government spending increase leads to a decrease in intertemporal wealth, as taxes will be imposed in the future to pay for the present expansion. This negative income effect, leads agents to work more and consume less decreasing real wages and leading to a higher marginal product of capital. At this juncture there are two differing views. Whereas Ramey and Shapiro (1998), drawing from the insights of models such as the one by Baxter and King (1993) argue that this rise in the marginal product of capital increases investment, they recognise that in some cases investment falls. This view is taken as the standard one by Leeper, Traum and Walker (2011) due to the low probabilities they find of an increase in output throughout all of their models. It is not impossible that the differences in the described effects on investment may be due to the fact that Ramey and Shapiro (1998) describe models where government spending is financed, through non distortionary taxes, while Leeper Traum and Walker (2011) consider deficit financed government spending. Indeed, according to Christiano, Eichenbaum and Rebelo (2009), the borrowing needs of the government lead to a crowding out of private investment and a rise in interest rates [According to Traum and Yang (2010), this is not necessarily the case]. As a fall in consumption and investment offsets government spending, the RBC fiscal multiplier is likely to be negative. However, if investment actually increases, it is possible to find a positive multiplier. Otherwise, as discussed by Perotti (2011), for government expenditure to have positive effects on output, it must be “sufficiently persistent” as to overcome the falls in consumption and investment. The introduction of rigidities such as habit formation and investment adjustment costs are only able to slow down the effects of the Ricardian equivalence, leading to qualitatively similar results.
The New Keynesian models begin to bias the multiplier in the opposite direction, with the likelihood of a positive multiplier progressively increasing as the model incorporates sticky prices, sticky wages and non-savers but decreasing again once the economy is described in an internationally open environment. According to Leeper, Traum and Walker (2011), wage and price stickiness implies that firms cannot hire or fire workers as easily as before, further reducing the effect of the Ricardian equivalence. As before, if wages cannot adjust immediately, the intertemporal wealth loss will continue to cause an increase in the supply of labour. However in this case it will not be met in the market, due to wage stickiness, and thus it stems the substitution effect that lowers consumption and increases labour. Therefore, there is a smaller fall in consumption as agents are less able to increase their labour supply. Moreover, if prices are also sticky, firms will react to the fiscal stimulus by increasing production rather than the price, leading to a fall in mark-ups, which will be translated into a fall in price and an increase in real wages, which in turn will increase consumption. The introduction of non-savers compounds these effects and their proportion of the total population seems to be a determining parameter. Because these agents consume all of their wealth each period, they do not suffer the wealth loss imposed by future taxes on savers. If non-savers represent a substantial fraction of all agents, then it is likely that the aggregate substitution effect will be positive, rather than negative, leading to an increase in the wealth and in consumption. However, this economy behaves differently in an open environment, where government spending leads to an increase in imports and thus decreases the size of the fiscal multiplier once more. In this setting, a positive government spending shock, as it increases consumption, will lead to an increase in domestic prices, which will motivate consumers to prefer foreign goods rather than domestic ones. The literature considers five other interesting scenarios. The first one compounds the effect of the open economy on the size of the fiscal multiplier, by treating fiscal expenditure as a “traded good”, in the sense that its deployment is not necessarily contained to the domestic economy but might extend to the rest of the world. Another scenario considers the responsiveness of monetary policy. If the monetary policy maker is passive and the fiscal policy maker is active, then the first will accommodate government spending by ensuring that the current budget constraint is satisfied, along the lines of Woodford (2003) and (2011). Indeed, when monetary policy is said to “lean against the wind” by responding less rapidly and proportionately to rises in inflation, the positive fiscal multiplier bias returns, leading to a 100% probability of a positive multiplier, according to Leeper, Traum and Walker (2011). Moreover, Christiano, Eichenbaum and Rebelo (2009) argue that the fiscal multiplier increases further when the zero lower bound on interest rates is binding. It has also been pointed out by Canzoneri et al (2011) that the multiplier is sensitive to the business cycle, assuming values as high as 2 in recessions. Finally, Forni and Pisanni (2010), estimate a model with all of these New Keynesian features for the Euro-zone. They account for a single monetary policy maker and consider fiscal policy scenarios under which authorities impose the same measures at the same time (coordinated), or not (uncoordinated), and find multipliers close to 1 for the first case.
III – Empirical Debates and Controversies
The other two approaches to this question are empirically based and attempt to tackle the problem of endogeneity, between fiscal policy and economic growth. Although both find that unexpected government spending has a positive effect on output, the identifications of such unexpected expenditures differ. Moreover, whereas according to Ramey (2011.a), the natural experiment approach finds negative effects on consumption and the real wage, the VAR approach normally finds a positive effect on these variables. Therefore, the empirical debate is related to the theoretical one, with the Natural Experiment approach used as an argument to defend the neo-classical/RBC models and the VAR approach used to support the claims of the New Keynesian
i.Natural Experiment approach
The first empirical methodology, proposed by Robert Barro’s 1981 seminal article and elaborated and popularised by Valery Ramey over the last 10 years, solves this problem by identifying purely exogenous forms of fiscal expenditure similar to natural experiments. This approach isolates periods of exogenous fiscal shocks and studies the effect of these specific shocks. Traditionally, it has used military expenditure during foreign military interventions to perform this identification, as in Barro (1981), Ramey and Shapiro (1998), Ramey (2011.a) and Barro and Redlick (2011). It finds positive output multipliers, where the substitution effect dominates, with a fall in consumption and the real wage leading to an increase in investment in line with Baxter and King (1993)’s neo-classical model discussed previously . The advantage of this method is that military interventions are highly visible and thus fairly easy to identify and study.
For all these reasons, the use of this method has been circumscribed to the USA where Ramey and Shapiro (1998) and Ramey (2011.a) estimate a fiscal multiplier close to 1. Barro and Redlick (2011), using a similar model, only find a multiplier of 0.5. However, Fisher and Peters (2010) using stock markets as measures of anticipated policy find a multiplier superior to 1, while Nakamura and Steinsson (2011), applying a deconstructed version of this approach to the subnational State jurisdictions of the USA, find a similarly large fiscal multiplier. Other competing identification strategies have been proposed. One such alternative was offered by Acconcia, Corsetti and Simonelli (2011) who use the removal of funds due to suspicion of political involvement with the Mafia in Italian municipalities as an exogenous shock to estimate the size of the local fiscal multiplier. Using this identifying strategy, the authors find a fiscal multiplier ranging between 1.4 and 2.
However, this approach is not without criticisms. First, such military interventions must be expeditionary in nature in order not to cause endogenous shocks. Clearly, the destruction caused by domestic military warfare would endogenise military expenditure. Second, the issue of timing requires an understanding of when economic agents change their expectations of future spending. The first issue is solved by the fact that the USA has only been involved in military conflicts on foreign soil during the twentieth century. However, although the first problem can be solved for the USA, such a solution is not necessarily available for other countries. Therefore, this approach is extremely limiting in terms of the case studies it can be extended to and its results should be heeded with caution as it may not be possible to generalise them to all countries or to other forms of expenditure which are not so easy to identify.
ii.Vector Autoregression (VAR) Approach
The second empirical manner to pursue this question is to accept the intrinsic endogeneity of this question and embrace it empirically through the use of Vector Autoregressions (VAR). This approach, informed by the insights Sims (1980) and the monetary applications of Bernanke and Mihov (1998), was popularised by Blanchard and Perotti’s 2002 study, which has become one of the seminal articles of research into this field of inquiry. This method is much less limiting because it does not require the identification of such restrictive exogenous forms of expenditure. In doing this it allows for the use of data sets with longer time series and the study of different forms of expenditure. Therefore, it is much more easily transported to other economies than the USA, and so provides for more generalisations.
The VAR approach finds contemporaneous shocks in the error terms of each of the simultaneous equations by imposing a structure that describes them as linear relationships of one another, rather than treating them as simple statistical errors of the model. Its main insight is to incorporate the fact that government spending, due to the democratic legislative process, cannot respond the a shock on output in the same quarter (contemporaneously), a fact missed by the use of annual data. The simplest VAR methodology estimates a recursive VAR (RVAR), estimated through a Cholesky factorisation of the residuals. The second approach is the structural VAR approach (SVAR), which elaborates on the RVAR approach by imposing a more sophisticated structural relationship on the residuals to identify the system as correctly as possible. Different models calculate parameter mixes in different manners and obtain different structures. Models can be just identified, as is the case in Fragetta and Melina (2010), or overidentified, as in Burriel et al 2009. Blanchard and Perotti’s (2002) find that a multiplier close to 1 for the USA from the first quarter of 1949 to the last quarter of 1997. Perotti (2005) and (2007) show that multipliers arrived at from the SVAR approach can also vary within a relatively large range, going from as low as -2.3 for Canada, to as high as 3.7 for the USA. Ilzetzki et al (2010) provides a similar study over a range of countries which allows him to control for specific economic features. In this study, the authors find that the level of development, fixed exchange rates closed economies and low levels of debt are associated with large fiscal multipliers. However, if the economies are open, developing, operating under flexible exchange rates and highly leveraged the fiscal multiplier is zero. Notwithstanding these country variations, estimates for the United States seem to be fairly consistent.
Of course the VAR approach is not without its criticisms. First, Ramey and Shapiro (1998) and Ramey (2011.a) note that not all the shocks identified can actually be said to be unanticipated. This is due to the fact that those models have traditionally tended not to include forecasts of government expenditure, which could otherwise provide more information to these models. This is a relevant criticism, the ignorance of which could potentially lead to a problem in causation. Edeblberg, Eichenbaum and Fisher (1999) also note that the identification of these disturbances is quite sensitive to the sample period, which makes the results unreliable. Recent research has sought to address these weaknesses. In Perotti (2011), the author finds that incorporating such forecasts in “expectations augmented” VAR (EVAR) yields “virtually the same results” as the standard SVAR approach. This conclusion is rebutted by Ramey (2011.b) who argues that the disagreement stems from a misunderstanding regarding the most recent methods used. Finally, Hall (2010) and Barro and Redlick (2011) argue that the limited variation in military data in the aftermath of the Korean war onwards invalidates any inferences from the SVAR, so the debate in this field is still ongoing.
Clearly the lines of division between the two general research approaches are fluid. For example, Edeblberg, Eichenbaum and Fisher (1999), Fisher and Peters (2010) and Ramey (2011.b) apply dummy variables to the SVAR approach to identify exogenous military expenditure shocks. Moreover, recent contributions to this faction of the dabate have incorporated other insights. Mountford and Uhlig (2009) estimate a Bayesian VAR to incorporate learning by economic agents and obtain positive multipliers for the USA. Auerbach and Gorodnichenko (2010) pursue a similarly sophisticated strategy to account for the role of interest rates and the business cycle and find similar results.
In conclusion, it is clear that the literature has evolved tremendously over the last twenty years, in all of its methodological forms. Having said this, the existing research and the heterogeneity of results it derives can seem confusing and contradictory. What the preceding lines have attempted to show is that this is not so. Rather, instead of being in turbulent conflict with each other a lot of the research is complementary and new layers add depth and qualification to the existing results. Despite the debate about the size of the fiscal multipliers, the dominant conclusion of the empirical research is that the effect is positive on output, as Figure 1 shows.
The main line of contention is not focused on the GDP multiplier but rather on the consumption and real wage multiplier. The natural experiment approach provides results that are coherent with a persistent government spending shock in a RBC framework. The new Keynesian approach incorporates an economy with frictions, habit prone agents, sticky prices and wages and non-savers in an internationally closed setting provides a description of the economy. Although Christiano et al (1999) and Vlaar (2004) make the closed economy assumption, Favero and Giavazzi (2008) argue that in light of the dominant effect of American monetary policy on long term government interest rate yields, it might be more advisable for the Euro area to be treated as a closed economy. The New Keynesian predictions are compatible with the empirical results found for the USA and Eurozone economies, in Burriel et al (2009) and in Ilzetzki et al (2010). Recent insights from Auerbach and Gorodnichenko (2010) also support the theoretical models that argue in favour of a multiplier that is sensitive to the interest rate zero lower bound and to the business cycle.
Burriel et al (2010)
Having ramble on to a rather substantial extent in the previous lines, I thought it would be relevant to show what fiscal multipliers actually look like in “real life”. To do this I use Burriel et al (2010), not because I endorse results (although they do make sense to me), but because his methodology and his approach are fairly representative of the standard of the empirical literature. In truth, they do not incorporate the forecasting insights described by Ramey and they are not as sophisticated as Auerbach and Gorodnichenko (2010)‘s use of the SVAR-DSGE tool kit. But they use a similar yardstick to the one developed by Blanchard and Perotti’s 2002, which they describe quite well. In particular they provide a comprehensive and accessible description of the structure that they impose to the SVAR’s error terms which can be replicated by anyone. Not only does it study the effect of fiscal multipliers in the USA, as usual, but also in the Euro-Zone, which makes it relevant for this blog.
The figures below show impulse response functions (IRFs) which are the coefficients used to estimate the effect of fiscal multipliers. This is a tool which was applied to economics by Helmut Lütkepohl.
Their results endorse the the mechanisms described in the theoretical section above. Government spending is procyclical, increasing GDP, although at a varying rate over time.
You might be interested to notice that according to estimates from the same study, taxes have procyclical effects, with constant albeit varying negative elasticities. So again, fiscal austerity, doesn’t really help economic growth.
Finally, having attempted to replicate their study, I can tell you that you can’t do it in Eviews because they overidentify the system of structural errors. Because SVARs run on Eviews automatically automatically use Maximum Likelihood to estimate the remaining coefficients in the structure, the results don’t make sense due to that overidentification. You may want to try RATS or R, both of which will allow you to use Instrumental Variables (IV) methods to estimate the coefficients of the structural errors.
I hope that this discussion has helped to clarify the issue of fiscal multipliers and the effects of fiscal policy. At a time when politicians seems eager to show their commitment to fiscal soundness, while promoting growth, all within a very weak economic context, this discussion should have clarified that austerity is unlikely to be the way. Clearly, the size of fiscal multipliers are debatable, but their direction is not. Certainly not under the present economic circumstances, where interest rates are close to or actually zero and where a lot of the monetary base has eroded the amount of money stock available to private investors. The issue is particularly relevant in Europe where the insights above ought to inspire leaders to accelerate the process of fiscal integration. This is a necessary condition if countercyclical fiscal policy is to have the necessary effects. We need coordinated fiscal policy that can push the monetary policy makers to “lean against the wind”. Unfortunately, it is an inescapable truth that any successful cooperation is impossible in Europe without delegation. The European Commission is the only path through which such a strong tool can find its full expression. However for the central bankers to “lean against the wind”, there needs to be democratic legitimacy. Although, this is to an extent present, the European Parliament needs to gain much more salience before the democratic legitimacy of European policies can be strong enough to challenge the strong conservative mandate of the ECB. This is what we learn from the USA. However, until that lesson is learned by European leaders, every other imaginable path will be attempted, creating havok and destruction along the way. What a pity…