Impact of Temporary Fiscal Shocks on the Canadian Economy - PDF document

Impact of Temporary Fiscal Shocks on the Canadian Economy Jean-Philippe Cayen Hélène Desgagnés Bank of Canada Bank of Canada Preliminary version, April 2009 Abstract In this paper, we assess the e¤ects of temporary …scal shocks on the Canadian economy using the structural vector autoregression ap- proach. We consider distinct shocks to government spending and tax revenues, and we confront three types of identi…cation approaches. The …rst one is the recursive approach based on the Cholesky decom- position. The second approach follows Blanchard and Perotti (2002) and Perotti (2004) who employed elasticities estimated using informa- tion on the tax system to identify the VAR model. In the last one, we impose restrictions on the sign of the variables’ responses to the shocks along the lines of Mountford and Uhlig (2005). We …nd that the e¤ects of the government revenues shock are more robust across the identi…cation approaches than the shock to government expendi- tures. For all the identi…cation approaches, the shock to government expenditures has a bigger impact on GDP than the net tax revenues shock in the very short run. However, on a horizon of two years, the revenues shock dominates because it has a larger stimulating e¤ect on the other components of GDP. In light of our …ndings, we view a speci…cation of the Blanchard and Perotti (2002) model in which the zero output elasticity of government expenditures assumption is dropped as the one leading to the most credible results. JEL classi…cation: E60, E62, H20, H50 Bank classi…cation: ... The views expressed in this paper are those of the authors. No responsibility for them should be attributed to the Bank of Canada.

1 Introduction The vector autoregression (VAR) methodology has become an important empirical tool for studying the e¤ects of …scal policy in recent years. For in- stance, the structural VAR model developed by Blanchard and Perotti (2002) is one of the most cited works on the topic. One of the major issues when working with a VAR model is to choose an appropriate approach to identify the structural shocks. To study the e¤ects of …scal policy using a VAR, we have to make the distinction between random discretionary …scal shocks and …scal policies that respond to the economic situation in a systematic way, such as the employment insurance bene…ts or the taxes on revenue. Besides the Blanchard-Perotti approach, other identi…cation strategies have been proposed to disentangle these two aspects of …scal policy such as the recursive approach (Fatas and Mihov, 2001) and the sign restrictions approach (Mountford and Uhlig, 2005). Unfortunately, all approaches do not yield the same results. For example, while most studies agree that a positive government spending shock will have a positive impact on GDP in the United States (see Blanchard and Perotti, 2002), this e¤ect is not always signi…cantly di¤erent from zero (see Perotti, 2004). There is not even a consensus whether an unexpetected tax cut will have a positive or a negative impact on the economy (see Perotti, 2004). Likewise, most of the empirical work on …scal policy has been done with U.S. data. Perotti (2004), and Phaneuf and Wasmer (2005) are among the few who have dealt with Canadian data in the VAR framework. The goal of our paper is to assess the e¤ects of temporary government spending and revenues shocks on the Canadian economy. For this purpose, we construct a VAR model and we compare the results we obtain using three identi…cation approaches: the recursive approach, the Blanchard-Perotti ap- proach and the sign restrictions approach. We …nd that the e¤ects of the government revenues shock are more robust across the di¤erent identi…cation approaches than the shock to government expenditures. An unexpected decrease of net tax revenues causes a grad- ual increase in real output, no matter which approach we use. The results are less consistent following the government spending shock. With the sign restrictions approach, a positive shock to government expenditures will be followed by a sustained increase of real GDP, while with the recursive and the standard Blanchard-Perotti approaches, the same shock generates a modest increase of output that lasts only one quarter. 1

In the next section, we brie‡y review previous work using various VAR models to study the e¤ects of …scal policies. Then, we describe our VAR model as well as the di¤erent approaches we test. In the last two sections, we present our results and discuss di¤erent robustness issues. 2 Literature Review Many papers deal with the impacts of …scal policies using VAR models. In most papers, the authors distinguish the impact of spending and revenues shocks instead of working only with the budgetary balance. This is probably explained by the fact that we expect the impacts to be di¤erent whether the government decides to cut the taxes or to increases spending. The recursive approach presented in Fatas and Mihov (2001), and Gali et al. (2007) restricts the matrix linking the reduced-form residuals to the structural shocks to be lower triangular to achieve identi…cation. As a result, the order of the variables in the model is crucial. Both papers …nd that the spending shock has a positive and signi…cant e¤ect on the level of output. Fatas and Mihov …nd a maximal multiplier impact of three. To reduce the number of zero restrictions imposed by the recursive strat- egy, Blanchard and Perotti (2002) incorporate external information on the tax system in their VAR to identify the shocks. They also …nd that the spending shock has a positive and signi…cant impact on the U.S. economy. On the contrary, a positive tax shock has a negative cumulative e¤ect on GDP. The sign restriction approach has been popularized by Uhlig (2005) who implemented it to evaluate the e¤ects of monetary policy shocks. Mountford and Uhlig (2005) use it to assess the e¤ects of …scal policy shocks on the U.S. economy. The idea is to restrain the direction of the responses to a speci…c shock in a way that is coherent with economic theory. It is somehow less restrictive than the two previous approaches since it does not impose any speci…c value for the models’s coe¢cients. On the other hand, imposing a signi…cant positive (negative) response of variables for many quarters could lead to upward (or downward) biased estimates. This approach leads to slightly di¤erent results than the recursive and the Blanchard and Perotti approaches. The rise in GDP following the spending shock is smaller than in previous papers. On the other hand, the tax shock as a signi…cant impact on GDP as Blanchard and Perotti (2002) …nd. 2

Since the results for the U.S. economy di¤er across the di¤erent papers cited above, Caldara and Kamps (2008) revisit all approaches to investigate whether the di¤erences are due to various reduce-form speci…cations or dif- ferent de…nitions of the data. They conduct their experiment with the same set of American data than Perotti (2004). They …nd that the spending shock has a positive impact on GDP. The response to a negative revenues shock is close to zero when they use the recursive or Blanchard-Perotti approaches, but positive with the sign restriction approach. Perotti (2004) extends the Blanchard-Perotti approach to …ve OECD countries - including Canada - and he …nds substantial di¤erences in esti- mated impacts of …scal shocks across countries. As a results, conclusions of previous work using U.S. data can hardly be transposed to Canada. Besides Perotti (2004), Phaneuf and Wasmer (2005) also use a VAR estimated with Canadian data. Perotti (2004) …nds a very modest response of GDP to the …scal shocks. Furthermore, the reponses depend on the subsample used to perform estimation. The estimated impact in Phaneuf and Wasmer (2005) is signi…cantly di¤erent from zero, but still modest. Alternative empirical tools have also been used to study the impact of …s- cal policies. As an example, Murchison and Robbins (2003) estimate an indi- cator of …scal policy stance by the generalized method of moments approach. This work is of a particular interest for us because they estimate the output elasticities of …scal variables, which can be used in a VAR model. However, their methodology is suitable for an assessment of permanent shocks. And, further, it does not bene…t from the dynamic framework of a VAR. 3 Methodology Similar to Blanchard and Perotti (2002) and to Phaneuf and Wasmer (2005), we use a three variables speci…cation for our base case VAR. Other authors like Fatas and Mihov (2001), Mountford and Uhlig (2002), Perotti (2004), Gali et al. (2007), and Caldara and Kamps (2008) use more variables in their model, but such a strategy generally requires the imposition of less realistic restrictions to identify the shocks. Nonetheless, we also present a sensitivity analysis that compares the results from alternative speci…cations to our base case results. The remainder of this section is divided into two parts. In the …rst one, we present the data used, while in the second part, we describe the three 3