Government expenditures have to be financed by taxes, eventually – a fact driven home by the current debate on US debt ceiling and the Eurozone sovereign-debt crisis.
The difficult political debate also shows that financing government expenditure is not trivial. If governments could easily increase their tax revenue without hampering economic activity, the debt ceiling debate would be void and Greece would not be receiving outside support. In reality, taxes distort economic decisions.
This simple insight carries through when analysing the costs and benefits of the fiscal stimulus in the aftermath of the recent financial crisis. Early analyses focused on quantifying the benefits of such a policy without modelling the government side in greater detail. Different policy rules to finance government debt make crucial differences when evaluating the recent stimulus.
This is not to deny that the financial crisis does make a difference for the policy analysis. It does. Theoretically, this has been shown by Eggertsson (2011) and Christiano et al. (2011). Simplifying:
- In the presence of the zero lower bound the nominal interest rate cannot be adjusted to accommodate an increased desire to save. Incomes have to fall.
- If the government creates inflation expectations by increasing its consumption, it can prevent such a fall.
If we analyse two economies, both with the same financial crisis but with and without fiscal intervention, the fiscal intervention causes output to rise, possibly dramatically.
With the potentially large appeal of government expenditure programmes in mind, it is therefore worthwhile to evaluate what policymakers in early 2009 could have believed about the effectiveness of the programme they eventually enacted, i.e. the American Recovery and Reinvestment Act (ARRA). In Drautzburg and Uhlig (2011), we undertake this analysis.
How to measure the effect of fiscal stimulus?
To measure the effect of the ARRA, we need a measure that compares the counterfactual path of output in its absence with the path of the economy when the ARRA is enacted and executed. Also, a summary statistic for the output effect is needed.
First, however, note that judging a policy by its output effects is not the same as a normative analysis. Raising output in the crisis need not increase the welfare of everyone to stabilise GDP in response to a crisis.
With this caveat in mind, consider how output effects should be measured. It is instructive to examine the components of the stimulus bill, which are plotted in Figure 1 as a percentage of GDP. Two features stand out. First, the expenditure is split unevenly over time. Second, only about a quarter of the stimulus is spent on government consumption.
To address the first feature of the data, spending and its output effect have to be aggregated over time. A natural choice is to discount costs and benefits by the discount rate the government faces on its debt. The ratio of the resulting present discounted values of the output effect relative to the cost of the stimulus is called the long-run multiplier. This statistic states how much GDP rises for each dollar spent on the ARRA.
Figure 1. Components of the American Recovery and Reinvestment Act, in % of 2008 GDP
To do justice to the different components of government spending, we proceed by extending standard macro models as they do not distinguish government consumption from government investment, and allow no role for distortionary taxes and transfer payments.
Which model to use?
A natural point of departure for the analysis of fiscal policy in the recent financial crisis is the medium-scale macro models, such as Smets and Wouters (2007), that are commonly used in the analysis of monetary policy. In such a model, the financial crisis can be modelled as either an exogenously fixed nominal interest rate or as an endogenously generated binding lower bound. Addressing the fiscal stimulus properly requires, however, a more detailed government sector, which is where our extensions come in.
Starting with the largest component of the stimulus plan, transfer payments play no role in these monetary macro models. Ricardian equivalence holds, meaning that issuing more debt to finance transfers today in exchange for a reduction in future transfers is irrelevant. Financing increased transfers today with distortionary taxes tomorrow would already break this equivalence. It would, however, not be in the spirit of the actual stimulus, which was meant to stimulate private sector consumption. A work-around is to introduce heterogeneous households, some of which consume their current period income and do not save, possibly because they are very impatient.
While government consumption is standard in monetary macro models, government investment is not. It does matter, however, and comprises between 3-4% of GDP. An intuitive way to model government capital is analogous to public infrastructure: it increases private sector productivity, but is subject to congestion. The government may then choose an optimal amount of investment to equate the discounted social marginal product of government capital to the cost. We can then consider how exogenous increases to government investment affect the economy.
Going back to the observation that in practice raising revenue for the government is costly, distortionary taxation should also be incorporated into the extended model. We follow Uhlig (2010), who considers distortionary taxes on consumption, hours worked, and capital. The government can issue debt, but eventually has to raise enough taxes to pay back the debt. In our baseline scenario, labour tax rates are adjusted.
Such an extended model features many parameters, only some of which are easily calibrated. Following the common practice in monetary macro models, we estimate a linearised version of the model using Bayesian techniques as a natural way to parameterise the model. In what follows, the results of this exercise, based on US post-war data up to the 4th quarter of 2008, are summarised.
When is fiscal stimulus effective?
Computing the effects of the stimulus plan based on the extended model yields the results shown in Figure 2. The figure illustrates the importance of our modelling choices.
First, the inclusion of the crisis with a binding bound on interest rates has an important effect on the multiplier. In early 2009, a reasonable guess was that interest rates would not be stuck at zero for more than eight quarters. Indeed, considering the prescribed interest rate path by the model and the historical conditions of 2009 confirms that durations between eight to twelve quarters were to be expected (“endogenous ZLB'' in the graph). In either case, the long-run multipliers are slightly negative, suggesting that apart from each dollar spent on the stimulus, the long-run financing does impose additional costs, although the ARRA does stimulate output over shorter horizons.
Figure 2.Multiplier in different scenarios, posterior median
The second feature illustrated in Figure 2 is how household heterogeneity matters. When transfers would be given to the Ricardian households exclusively, the output effect of the stimulus would be much more benign as the Ricardian households adjust their savings decisions. Yet, multipliers eventually turn negative even when transfers also go to the households who consume their current income.
Financing schemes matter
To understand why the long-run output effect of the stimulus is negative, the financing scheme is crucial. Consider lowering lump-sum transfer payments to households to finance the stimulus or raising consumption taxes. Adjusting transfers does not affect Ricardian households directly, whereas adjusting consumption or labour tax rates distorts their choices. Figure 3 shows that adjusting transfers has no negative long-run effects – the long-run multiplier in the right panel of Figure 3 is similar to that at a one year horizon in the left panel. It is indeed slightly higher as part of the stimulus takes effect after one year. In contrast, when financing the expenditure via labour taxes (as in the baseline case) or consumption taxes, the long-run effects become negative.
Figure 3. Short and long-run multipliers with an 8-quarter ZLB under different financing schemes – posterior distribution
Figure 4.Effects of broadening the labour tax base on the multiplier with a ZLB duration of eight quarters, posterior distribution
Figure 4 illustrates a related implication, i.e. the well-known fact that if distortionary taxes have to be used, their effect is more benign if the tax base is broader. In the case of distortionary labour tax, the tax base is inversely related to the capital share. When the capital share is increased, the labour tax rate has to rise more. This amplifies the negative long-run effects.
In summary, abstracting from financing decisions in the evaluation of the stimulus programme gives a wrong impression of the net benefits of stimulus. Government debt has to be paid back sooner or later and the fiscal consolidation is costly for the economy.
Christiano, Lawrence, Martin Eichenbaum, and Sergio Rebelo (2011), “When Is the Government Spending Multiplier Large?”, Journal of Political Economy, 119(1):78-121.
Drautzburg, Thorsten and Harald Uhlig (2011), “Fiscal Stimulus and Distortionary Taxation”,NBER Working Paper 17111.
Eggertsson, Gauti B (2011), “What Fiscal Policy is Effective at Zero Interest Rates?”NBER Macroconomics Annual 2010 (25):59-112.
Smets, Frank and Rafael Wouters (2007), “Shocks and Frictions in US Business Cycles: A Bayesian DSGE Approach”, American Economic Review, 97(3):586-606.
Uhlig, Harald (2010), "Some Fiscal Calculus", American Economic Review, 100(2):30-34.