Abstract
This Selected Issues paper analyzes the performance of French banks and the financial support measures taken by the French government. France has a large and sophisticated financial system, which accounts for 10 percent of the global banking system and 5 percent of global capital markets. This paper presents an overview of the French banking and supervision structure. It conducts comparative analyses of profitability, asset quality, capital adequacy, leverage, quality of capital, funding profile, and liquidity of banks. The paper also analyzes business lines, potential spillovers, and market perceptions of risk.
II. Countercyclical Stimulus and Long-Term Sustainability: Insights From a Fiscal SVAR for France4
A. Introduction
46. At this time when fiscal policy instruments are called into action, it is useful to take stock of their effectiveness. This chapter presents new estimates of fiscal multipliers for France, while comparing those with earlier estimates obtained for France and comparator countries. Estimates of fiscal multipliers can help shed some light on two pressing policy questions: what is the output impact of countercyclical fiscal policy? And when the recovery is underway, how best to implement the necessary fiscal consolidation?
47. This chapter uses a Structural Vector Auto Regression (SVAR) model to estimate fiscal multipliers and derive policy recommendations for countercyclical policy and for preserving long-term sustainability. In a seminal paper, Blanchard and Perotti (2002) propose a methodology to identify fiscal policy shocks which allows to derive stylized facts about their effectiveness. The framework has since been extended by Perotti (2002) and Favero and Giavazzi (2007). In particular, the latter paper explicitly integrates a debt feedback rule into the model. I apply this recent methodology to French data over the past thirty years. As the VAR methodology is essentially backward-looking, giving an average response of the economy to a fiscal shock, conditional of the specific environment prevailing during the sample period, I also discuss whether fiscal multipliers might have changed and to what extend historical estimates of multipliers could be modified in the current environment.
48. The chapter is organized as follows. Section B presents the methodology and the data used. Section C presents estimates of fiscal multipliers in a model with or without debt feedback. Section D discusses the results, and section E concludes.
B. Methodology and Data
Theoretical setup
49. Two broad set of methodologies have been used to assess the impact of fiscal policy shocks. One relies on structural macroeconomic models, the other on econometric estimations. Both have specific advantages and drawbacks, one often the mirror image of the other. Structural macroeconomic models are founded on a well-specified theoretical framework and, depending on the level of detail of the model, allow to capture the impact of very specific shocks (e.g. decrease of social security contribution on wages) and to follow the propagation of the shock throughout the variables of interest included in the model. However, despite efforts at calibrating accurately the models, they may rely on theoretical assumptions that are too strong and not validated by the data. While the broad results of structural macroeconomic models are similar, a comparison of the results of models used in different international organizations show occasionally some significant differences (see IMF, 2008; and Laxton, 2009).5
50. Econometric models allow to capture the average historical response to the impact of specific shocks. In contrast to macroeconometric models, econometric methods, such as VAR, do no rely so strongly on theoretical assumptions but allegedly let the data “speak for itself.” One main difficulty is the correct identification of the shocks. Here also, two main approaches can be distinguished. The first relies on event studies, which aim at capturing pure exogenous shocks, either for expenditure (such as war build-ups, as in Ramey and Shapiro, 1998) or for revenue (regulatory changes of tax policy, as in Romer and Romer, 2007). This may require constructing a very detailed and time-intensive information set on the chronology and the size of the shocks, but does not completely avoid the issue of exogeneity. The second approach consists in deriving the policy shocks as the residuals of a VAR model, and imposes a structural interpretation to the residuals to avoid their autocorrelation. Building on earlier models applied to monetary policy, Blanchard and Perotti (2002) have pioneered the extension of the SVAR methodology to fiscal policy. Whether this method truly captures reduced-form shocks is the object of a lively debate between the proponents of the two approaches. Both econometric methods have the main drawback of being backward-looking, namely reflecting the average economic conditions over the period of estimation.
51. The originality of the Blanchard-Perotti SVAR model is to use institutional information to impose a structural interpretation on the residuals. They estimate a VAR with only three variables (government revenues, government expenditures, output). Standard residuals of the VAR are correlated, because of the effect of output on revenues—the “automatic stabilizer” effect—and the opposite impact of revenues on output—the multiplier impact that the model attempts to measure. Using institutional information on the spontaneous reaction of revenues to an output shock and under the assumption of no reaction of expenditures to an output shock within the same quarter (reflecting institutional delays in deciding or implementing a policy response) allows to construct reduced-form uncorrelated residuals.
52. The basic specification is the following. Let Yt be the three-variable VAR (with revenues tt, expenditures gt, and output xt - with all variables in logarithms), and k the number of lags in the VAR:
where Ut is the vector of standard residuals of the VAR. The methodology consists in finding two matrices, A and B, such that:
in which Et is a vector of uncorrelated structural residuals.
53. The construction of the structural residuals relies on a combination of institutional information and assumptions. Specifically, it relies on the construction or estimation of a number of parameters such that:
For example, the spontaneous response of revenues to output ηTY uses information on the elasticity within the same quarter of each tax base to output shock, ηBiY, and of the tax revenue itself to the tax base, ηTBi. The response of the aggregate revenues to an output shock is then constructed as a weighted average:
54. Successive SVAR models applied to the analysis of fiscal policy have allowed to gradually broaden the original model. Perotti (2002) extends the original methodology to encompass price and interest rate feedbacks. He uses similarly institutional information to estimate reduced form residuals for the two additional variables. For example, the contemporaneous response of social benefits in real terms to price is assumed to be -1, since social benefits are usually not indexed to inflation within the same quarter.
55. One criticism of standard fiscal SVAR models is that they fail to account for debt dynamics. Blanchard and Perotti acknowledge that it is one of the two “crimes” committed in their paper (the second being ignoring the Lucas critique). Favero and Giavazzi (2007) argue that traditional SVAR fail to keep track of the consequent debt developments and overlook the possibility that fiscal variables might respond to the debt variable, as the debt ratio evolves over time. As a consequence, the error terms captured by the SVAR include not only the exogenous fiscal shocks but also the responses of taxes, spending, and long-term interest rates to a debt shock. In the empirical part of their paper, they find that introducing debt dynamics tends to reduce the size of multipliers.6
56. In practice, the Favero-Giavazzi method relies on estimating feedback effects of debt while introducing a debt accumulation equation. Debt variables are included in the VAR while an additional accounting equation is added to close the model and to derive impulse responses for the VAR. Specifically, equation (1) is modified as follows:
where dt is the debt-to-GDP ratio and g't is the logarithm of primary expenditures. The impulse response function is computed differently than in a standard VAR. It requires first creating a baseline by solving the model dynamically forward without shock. Then, each specific variable of the VAR is subject to a shock, and the model is again solved dynamically forward. The impulse response is deduced from the difference between both scenarios. The dynamics in play here is non-linear as it depends on the initial level of debt.
57. A broad set of results using traditional fiscal SVAR models is already available. Most of the fiscal SVAR methodology has been developed and applied to data for the United States for the past forty-fifty years, which allows for a better comparability of the results (Blanchard-Perotti, 2002; Perotti, 2002; or Favero-Giavazzi, 2007). Perotti (2002) extends the methodology to five countries, including three European countries. Specific studies using the same methodologies have also been done for a number of other European countries. On France, Biau and Girard (2006) applies the SVAR methodology in a VAR with five variables as in Perotti (2002). Their paper presents, however, two marginal differences with the rest of the literature: due to data availability, the sample period is shorter, starting only at the end of the 1970s; and the main aggregates are differentiated, thus loosing some possible information from the level of the variables. In another VAR setup for France, Boissinot, L’Angevin, and Monfort (2004) introduce a debt feedback, but as the focus of their paper is more on the debt dynamics than on the measure of fiscal multipliers, a simpler Cholesky ordering is applied to the residuals rather than the more elaborate structural interpretation à la Blanchard-Perotti.
Data
58. The sample period covers the past 30 years. The data used are quarterly data spanning over 1978Q1–2007Q4. To the five original variables of Perotti (2002), i.e. real GDP, government expenditures, government revenues, inflation proxied by the change of the GDP deflator, and interest rate—is added a debt variable. National accounts data from INSEE is complemented by data from the quarterly database of the OECD. In particular, government expenditures on goods and services come from INSEE but the remaining fiscal data comes from the OECD. GDP and fiscal aggregates are deflated by the GDP deflator. The debt variable is the gross public debt consistent with the Maastricht definition. As quarterly debt data are not available before 1995, the chapter uses the methodology exposed in Boissinot, L’Angevin, and Monfort (2004) to construct a quarterly debt stock for the sample period. Annex Figure 1 presents the main variable of interest.
59. The definition of the variables is tailored to the estimation strategy. The
definition of fiscal variables varies significantly across different papers. Most the SVAR literature on the impact of fiscal policy focuses on government expenditures on goods and services on one side, and on taxes net on transfers on the other (Blanchard-Perotti, 2001; or Biau-Girard, 2005). In Favero-Giavazzi (2007), by contrast, the definition of the fiscal variables is determined by their need to construct a debt accumulation equation. They thus focus on primary current expenditures and on net taxes. This consideration leads them also to focus on the implicit interest rate on government debt, rather than on the monetary policy rate.7 I follow the same methodology as Favero-Giavazzi (2007). The robustness of the results to alternative definitions of the fiscal variables is also discussed.
Model specification
60. As the variables of interest are non-stationary, a cointegrated VAR setup is considered. The time series of the chapter are integrated of order one in levels, but stationary in differences (see Annex Table 1). Thus a cointegrated system seems the most appropriate approach to estimate the SVAR, without discarding valuable information present in the level of the variables. Blanchard and Perotti (2002) estimate their SVAR model either with a cointegration vector or after detrending the time series. Biau and Girard (2006) estimate a VAR in differences, as they argue that it is difficult to provide a meaningful economic interpretation to the cointegration vectors present in the econometric system. While there is strong suggestion that most other studies use a SVAR model in levels, the exact treatment applied to the non-stationarity of the variables is often unclear. For robustness and for comparability with the Biau-Girard paper, results from a SVAR model in differences are also discussed.
61. The specification of the VAR is determined using traditional tests. The lag
structure of the VAR is determined by information criteria (AIC, BIC). In the preferred specification, four lags of the VAR are considered. The number of cointegration relations is determined by the trace and eigenvalue tests. The specification of the SVAR is not affected by the inclusion of the debt variable on the right hand side as in equation (4). In the preferred specification, four cointegrated relations are retained. No specific constraints are imposed on the form of the cointegration vector. The number of cointegration vectors seems large, since one could expect to find at least two stochastic trends, one for the drift of real GDP reflecting productivity and factor accumulation, and the other for the trend decline of the inflation rate over the period considered. Eyeballing the data suggests a number of candidates for the cointegration relations, notably the real interest rate, broadly constant over the period; the fiscal deficit or some weaker form of relationship between expenditures and revenues; and the broad stability of revenue or expenditure-to-GDP ratios over the period considered.
62. Structural residuals seem to adequately capture the main fiscal policy changes over the sample period. Annex Figure 2 presents the structural residuals of the SVAR and indicates notable fiscal policy changes using the chronology established by Biau-Girard. They show that the largest residuals generally capture policy changes. In addition, the structural residuals are broadly similar in the models with or without debt feedback.
C. Fiscal Multipliers in a Model With or Without Debt Feedback
Baseline fiscal multipliers
63. Estimates in a model without debt feedback give a peak multiplier of 1.1 for expenditures and 0.3 for revenues.8 The expenditure multiplier rises gradually from 0.4 after a year to 0.6 after two years and to a peak value of 1.1 after six years before declining slightly. The revenue multiplier peaks after eight quarters around 0.3, then declines to a long-term value of 0.2. During the first year, the sign of the multiplier is positive, implying that a tax increase induces an increase in GDP. While the hierarchy between revenue and expenditure multipliers is somehow comparable to that obtained in other studies, other results are surprising, e.g. the extremely gradual rise of the expenditure multiplier, suggesting long delays for the negative feedbacks from interest rate and prices to kick-in; the non-standard value of the revenue multiplier during the first year; and the undulation of the revenue multiplier over time. Note also that the peak revenue multiplier is significantly different from zero at the 10 percent level only around eight quarters. By contrast, after six quarters, the expenditure multiplier is significantly different from zero at the 10 percent level.
Impulse Response Functions to Fiscal Shocks in Baseline Model
Citation: IMF Staff Country Reports 2009, 233; 10.5089/9781451813739.002.A002
Source: Fund staff estimates.NB: Dotted lines represent the confidence intervals at 10 percent; the impulse responses present the impact of a temporary one unit shock of the fiscal variables on the level of output.Impulse Response Functions to Fiscal Shocks in Baseline Model
Citation: IMF Staff Country Reports 2009, 233; 10.5089/9781451813739.002.A002
Source: Fund staff estimates.NB: Dotted lines represent the confidence intervals at 10 percent; the impulse responses present the impact of a temporary one unit shock of the fiscal variables on the level of output.Impulse Response Functions to Fiscal Shocks in Baseline Model
Citation: IMF Staff Country Reports 2009, 233; 10.5089/9781451813739.002.A002
Source: Fund staff estimates.NB: Dotted lines represent the confidence intervals at 10 percent; the impulse responses present the impact of a temporary one unit shock of the fiscal variables on the level of output.64. The results are broadly robust under alternative assumptions. Peak multipliers are broadly similar when using different assumptions, such as restricting the model to the same sample as Biau-Girard i.e. 1978–2003; when using the fiscal variables and GDP in growth rates instead of levels, again as Biau-Girard; or when using only a SVAR with three variables as in the Blanchard-Perotti paper, in which case the peak tax multiplier is slightly higher at 0.5.
65. Recursive estimation also shows a broad stability of the multipliers over the past five years. One preliminary way to estimate, within a traditional SVAR model, the impact of the debt level is to see whether the long-run multipliers are affected by the estimation period. The short sample available prevents from estimating the model over different sub-samples. One could, for example, expect the multiplier to decrease when more recent data (with the debt ratio hovering over 60 percent since the early 2000s) is included. However, recursive estimations (by gradually shrinking the sample period) show that model estimates are broadly similar whether the past five years are included or not. By contrast, estimates change drastically when the model is only estimated over 1978–2001, with the tax multiplier being higher but the expenditure multiplier being lower. Given the contrasting evolution of the tax and expenditure multipliers, we interpret this more as evidence of the poor statistical properties of the model over a reduced sample, than as a change of the estimates because of the higher debt level.
Recursive Estimation of Long-Term Fiscal Multipliers
Citation: IMF Staff Country Reports 2009, 233; 10.5089/9781451813739.002.A002
Source: Fund staff estimates.Recursive Estimation of Long-Term Fiscal Multipliers
Citation: IMF Staff Country Reports 2009, 233; 10.5089/9781451813739.002.A002
Source: Fund staff estimates.Recursive Estimation of Long-Term Fiscal Multipliers
Citation: IMF Staff Country Reports 2009, 233; 10.5089/9781451813739.002.A002
Source: Fund staff estimates.Fiscal multipliers with debt feedback
66. Introducing a debt feedback leads to broadly similar results in the short run but to strikingly different results in the long run. In the short term, the results are similar to those obtained in the Perotti (2002) setup. Tax multipliers are lower than expenditure multiplier (0.2 vs. 0.8). By contrast, expenditure multipliers become negative, then null. Tax multipliers turn negative after eight quarters and remain negative in the long run (consistent with an interpretation in terms of expansionary fiscal consolidation).
Multipliers in Baseline Model (dotted lines) and Model with Debt Feedback (solid lines)
Citation: IMF Staff Country Reports 2009, 233; 10.5089/9781451813739.002.A002
Source: Fund staff estimates.Multipliers in Baseline Model (dotted lines) and Model with Debt Feedback (solid lines)
Citation: IMF Staff Country Reports 2009, 233; 10.5089/9781451813739.002.A002
Source: Fund staff estimates.Multipliers in Baseline Model (dotted lines) and Model with Debt Feedback (solid lines)
Citation: IMF Staff Country Reports 2009, 233; 10.5089/9781451813739.002.A002
Source: Fund staff estimates.67. The results of the model with debt feedback need to be taken with caution for a number of reasons:
The contrast between the results of both models are sharper than the difference obtained for the United States by Favero-Giavazzi (2007), where the efficiency of fiscal instruments is only marginally reduced but not cancelled or reversed.
The coefficients of debt in some of the equations of the SVAR display occasionally an unexpected sign; for example, it is possible that the negative sign of the debt variable in the interest rate equation may capture the historical combination of the decline of nominal interest rate while debt increases.
The impulse response functions present a significant cyclicality which is difficult to explain and may reflect some misspecifications.
In general, as illustrated in Table 1 below, SVAR results may vary significantly from one study to the next, depending on the time period or the variables considered.
68. However, one result that stands out across models and specifications is the hierarchy between expenditure and revenue multipliers. The long-term expenditure multiplier for France found here in the model without debt is on the high side compared to a sample of other G7 countries, but lower than the value found by Biau-Girard (2005). By contrast the long-term expenditure multipliers in the model with debt are comparable to results found by different authors for Italy, Spain or the U.K. The expenditure multiplier based on the Ramey and Shapiro (1998) event study for the U.S. tends to be higher in the short term but lower in the long term, but it is uncertain whether this result can be generalized given the specificity of the shocks studied (mostly military expenditures). In the baseline model, the peak revenue multiplier is higher that in most other studies. One exception, are the results of Romer and Romer (2007), based not on a SVAR model, but on event study, and which found a much larger revenue multiplier.
Survey of Fiscal Multipliers 1/
The table presents the impact of an expensionary fiscal shock of one unit (expenditure increase, revenue decrease).
Impact on private GDP.
Military spending only.
Survey of Fiscal Multipliers 1/
| Country | Author (s) | Period | Expenditure | Revenue | ||||||
|---|---|---|---|---|---|---|---|---|---|---|
| One year | Peak | Long term | One year | Peak | Long term | |||||
| SVAR models | ||||||||||
| France | This study | w/o debt | 1978 | 2007 | 0.62 | 1.11 | 1.08 | 0.07 | 0.37 | 0.21 |
| with debt | 1978 | 2007 | 0.40 | 0.80 | -0.10 | 0.00 | 0.20 | -0.80 | ||
| France | Biau Girard, 2005 | 1978 | 2003 | 1.40 | 2.00 | 0.80 | 0.10 | … | … | |
| U.S.A. | Blanchard-Perotti, 2002 | 1947 | 1997 | 0.85 | 1.41 | 0.69 | 0.64 | 0.71 | -0.11 | |
| U.S.A. | Perotti, 2002 | 1961 | 2000 | 0.29 | 1.05 | 0.96 | 0.66 | 0.75 | 0.53 | |
| U.S.A. | Favero-Giavazzi, 2007 | w/o debt | 1980 | 2006 | 0.16 | 0.80 | 1.43 | 0.00 | 0.17 | 0.27 |
| with debt | 1980 | 2006 | 0.13 | 0.71 | 1.28 | -0.01 | 0.28 | 0.65 | ||
| Germany | Perotti, 2002 | 1961 | 1989 | 0.96 | 1.30 | 0.94 | 0.46 | 0.98 | -0.05 | |
| Germany | Heppke-Falk et al., 2006 | 1974 | 2004 | 0.62 | 1.23 | 0.38 | -0.08 | -1.51 | -1.51 | |
| UK | Perotti, 2002 | 1964 | 2001 | -0.04 | 0.30 | -0.06 | -0.05 | -0.21 | -0.24 | |
| Italy 2/ | Giordano et al., 2007 | 1982 | 2004 | 0.25 | 0.50 | -0.25 | 0.08 | 0.16 | 0.05 | |
| Spain | Castro et al., 2007 | 1994 | 2004 | 0.10 | 0.20 | -0.10 | 0.00 | 0.04 | -0.10 | |
| Event studies | ||||||||||
| U.S. | Romer-Romer, 2008 | 1.20 | 2.80 | 2.70 | ||||||
| U.S. 3/ | Ramey-Shapiro, 2002 | 1.50 | … | -0.20 | ||||||
| Macroeconometric models | ||||||||||
| France | Mesange, 2001 | 1.30 | … | … | 0.45 | … | … | |||
| France | GIMF, 2009 | 1.06 | … | … | 0.38 | … | … | |||
The table presents the impact of an expensionary fiscal shock of one unit (expenditure increase, revenue decrease).
Impact on private GDP.
Military spending only.
Survey of Fiscal Multipliers 1/
| Country | Author (s) | Period | Expenditure | Revenue | ||||||
|---|---|---|---|---|---|---|---|---|---|---|
| One year | Peak | Long term | One year | Peak | Long term | |||||
| SVAR models | ||||||||||
| France | This study | w/o debt | 1978 | 2007 | 0.62 | 1.11 | 1.08 | 0.07 | 0.37 | 0.21 |
| with debt | 1978 | 2007 | 0.40 | 0.80 | -0.10 | 0.00 | 0.20 | -0.80 | ||
| France | Biau Girard, 2005 | 1978 | 2003 | 1.40 | 2.00 | 0.80 | 0.10 | … | … | |
| U.S.A. | Blanchard-Perotti, 2002 | 1947 | 1997 | 0.85 | 1.41 | 0.69 | 0.64 | 0.71 | -0.11 | |
| U.S.A. | Perotti, 2002 | 1961 | 2000 | 0.29 | 1.05 | 0.96 | 0.66 | 0.75 | 0.53 | |
| U.S.A. | Favero-Giavazzi, 2007 | w/o debt | 1980 | 2006 | 0.16 | 0.80 | 1.43 | 0.00 | 0.17 | 0.27 |
| with debt | 1980 | 2006 | 0.13 | 0.71 | 1.28 | -0.01 | 0.28 | 0.65 | ||
| Germany | Perotti, 2002 | 1961 | 1989 | 0.96 | 1.30 | 0.94 | 0.46 | 0.98 | -0.05 | |
| Germany | Heppke-Falk et al., 2006 | 1974 | 2004 | 0.62 | 1.23 | 0.38 | -0.08 | -1.51 | -1.51 | |
| UK | Perotti, 2002 | 1964 | 2001 | -0.04 | 0.30 | -0.06 | -0.05 | -0.21 | -0.24 | |
| Italy 2/ | Giordano et al., 2007 | 1982 | 2004 | 0.25 | 0.50 | -0.25 | 0.08 | 0.16 | 0.05 | |
| Spain | Castro et al., 2007 | 1994 | 2004 | 0.10 | 0.20 | -0.10 | 0.00 | 0.04 | -0.10 | |
| Event studies | ||||||||||
| U.S. | Romer-Romer, 2008 | 1.20 | 2.80 | 2.70 | ||||||
| U.S. 3/ | Ramey-Shapiro, 2002 | 1.50 | … | -0.20 | ||||||
| Macroeconometric models | ||||||||||
| France | Mesange, 2001 | 1.30 | … | … | 0.45 | … | … | |||
| France | GIMF, 2009 | 1.06 | … | … | 0.38 | … | … | |||
The table presents the impact of an expensionary fiscal shock of one unit (expenditure increase, revenue decrease).
Impact on private GDP.
Military spending only.
D. Discussion of the Results
69. The results suggest consistently that the most effective way to stimulate the economy is through expenditure increase rather than through a tax cut. The higher “bang for the buck” in the short term of an expenditure increase suggests that this should be the preferred instrument to inject a countercyclical stimulus. This hierarchy is maintained also in the long run in both models. The model without debt suggests a lasting impact of an expenditure increase (1.08) against a smaller one for a tax cut (0.20). The model with debt suggests that an expenditure increase is almost neutral (-0.10) while a tax cut would have a significant negative output impact (-0.80), possibly because of Ricardian effects.
70. The model also illustrates some of the lessons derived by Kumar, Leigh, and Plekhanov (2007) from numerous experiences of fiscal consolidation across the world. For example, they find that fiscal consolidations can be expansionary provided they do not rely on cutting productive public expenditures. The models here also suggest that expenditure cuts could be harmful in the short run, while being mildly expansionary in the long run in the setup that fully integrates the impact of debt feedback. They also conclude that most consolidations were launched during economic downturns or the early stages of recovery. Provided the consolidation strategy rely on a low “bang for the buck,” it would indeed avoid hurting the recovery while delivering high fiscal payoffs. For example, raising some taxes (which have a lower tax multiplier) would have little negative impact on growth. Prime candidates for consolidation seem France’s large tax expenditures, which should be evaluated both from an economic efficiency and equity point of view. Beyond that, however, there is little scope for consolidation through tax hikes, given the already high tax ratio. In addition, historical experience shows that fiscal adjustments that rely on cuts in current expenditures tend to be more durable than revenue-based consolidations.9
71. The impulse response of the VAR also suggests revenue and expenditure shocks are highly persistent, with revenue shocks possibly slightly more so. In both models, a one-off increase in revenue by 1 percentage point of GDP leads to a long-term increase of 1.2 percentage point. By contrast, expenditure shocks seem to be less persistent but this depends on the model: in the model without debt, about 4/5 percent of a one-off expenditures shock remains in the long run, against only 1/5 percent in the model with debt feedback. The model with debt also suggests that any increase in expenditure tends to give rise to an offsetting increase of revenues, but while the additional expenditures are broadly matched in the medium run, in the long run only a third of the additional expenditures are covered by new revenues.
Impulse Responses of Fiscal Shocks on Themselves in Baseline (dotted lines) and Extended Models (plain lines)
Citation: IMF Staff Country Reports 2009, 233; 10.5089/9781451813739.002.A002
Source: Fund staff estimates.Impulse Responses of Fiscal Shocks on Themselves in Baseline (dotted lines) and Extended Models (plain lines)
Citation: IMF Staff Country Reports 2009, 233; 10.5089/9781451813739.002.A002
Source: Fund staff estimates.Impulse Responses of Fiscal Shocks on Themselves in Baseline (dotted lines) and Extended Models (plain lines)
Citation: IMF Staff Country Reports 2009, 233; 10.5089/9781451813739.002.A002
Source: Fund staff estimates.72. How valid are the multipliers estimated in this chapter for the current crisis? There is a lively debate about whether historical fiscal multipliers as derived from econometric estimates are of any use to assess the impact of fiscal policy in the current crisis. Some expect multipliers to be larger because of the large unutilized capacity which would limit the subsequent crowding out of the private sector; because of the dysfunctionality of the credit market which constrains the availability of credit to households and thus raises their propensity to consume out of income; or because of a possible confidence effect of fiscal policy allowing the economy to jump from a low to a high confidence equilibrium. By contrast, multipliers could be lower if Ricardian effects dominate and doubts about the long-term sustainability of public debt raise household savings, while the poor state of household and corporate balance sheets would force them to consolidate. The results of the SVAR models suggest that these considerations are not necessarily mutually exclusive, since low or negative long-run multipliers may indicate Ricardian behavior, without cancelling out the positive short-term impact of countercyclical policy.
E. Conclusions
73. This chapter finds that introducing a debt equation reduces significantly the size of multipliers. In the baseline model without debt, spending multipliers are larger than tax multipliers, in line with comparator results. By contrast, when a debt equation is introduced, while short-term multipliers are not modified, long-term expenditure multipliers go to zero, and long-run tax multipliers turn negative. While the results with the debt equation need to be interpreted with caution, overall results suggest the following policy implications:
In the short run, a stimulus based on increasing expenditures is more efficient than one based on tax cuts.
Over the long run, however, and taking into account the debt feedback, a consolidation based on expenditures does not have a negative impact on activity. Negative long-run tax multipliers would suggest that considerations of fiscal sustainability may dominate the countercyclical impact of a tax cut.
References
Allard-Prigent, Celine, Cedric Audenis, Claire Berger, Nicolas Carnot, Sandrine Duchêne and Fabrice Pesin, 2002, “Présentation du modèle MESANGE,” Document de travail de la Direction de la Prévision.
Blanchard, Olivier, and Roberto Perotti, 2002, “An Empirical Characterization of the Dynamic Effects of Changes in Government Deficit and Taxes on Output,” Quarterly Journal of Economics, Vol. 117 (November) pp. 1329- 68.
Biau, Olivier, and Elie Girard, 2006 “L’apport d’un modèle VAR structurel,” Economie et Prévision.
Boissinot, Jean, Clotilde L’ Angevin, and Brieuc Monfort, 2004, “Public debt sustainability: some results for the French case,” Document de travail de l’INSEE.
Castro, Francisco de, Pablo Hernandes de Cos, “The Economic Effects of Fiscal Policy: The case of Spain,” Journal of Macroeconomics, 2007
Dai, Qiang and Philippon, Thomas, 2005, “Fiscal Policy and the Term Structure of Interest Rates,” NBER Working Paper no. W11574.
Favero, Carlo, and Francesco Giavazzi, 2007, “Debt and the Effects of Fiscal Policy,” NBER Working Paper no. 12822.
Kumar, Manmohan, Daniel Leigh, and Alexander Plekhanov, 2007, “Fiscal Adjustments: Determinants and Macroeconomic Consequences,” IMF Working Paper no. 07/178.
Heppke-Falk, Kirsten, Jorn Tenhofen, and Guntram Wolff, 2006, “The macroeconomic effects of exogenous fiscal policy shocks in Germany: a disaggregated SVAR analysis,” Deutsche Bundesbank, Discussion Paper #41.
IMF, 2008, “Fiscal Policy as a Countercyclical Tool,” World Economic Outlook October.
Laxton, Doug, 2009, “Effects of Fiscal Stimulus in Structural Models,” IMF, mimeo.
Perotti, Roberto, 2002, “Estimating the Effects of Fiscal Policy in OECD Countries,” ECB Working Paper.
Ramey, Valerie, and Matthew Shapiro, 1998, “Costly Capital Reallocation and the Effects of Government Spending,” Carnegie-Rochester Conference Series on Public Policy, Vol. 48 (June) pp. 145- 94.
Romer, Christina, and David Romer, 2007, “The Macroeconomic Effects of Tax Changes: Estimates Based on a New Measure of Fiscal Shocks,” NBER Working Paper no. 13264.
Yakadina, Irina, and Boriana Yontcheva, 2009, “Recession and Recovery: Automatic Stabilizers and Discretionary Fiscal response in France,” IMF, France: Selected Issues.
Annexes
Stationarity Tests and Order of Integration of the Variables
Stationarity Tests and Order of Integration of the Variables
| ADF test | KPSS | |||||
|---|---|---|---|---|---|---|
| Level | First diff. | Order integ. | Level | First diff. | Order integ. | |
| Real fiscal expenditures | -1.92 | -4.09 | I(1) | 0.26 | 0.06 | I(1) |
| Real fiscal revenues | -1.73 | -5.99 | I(1) | 0.10 | 0.06 | I(0) |
| Real GDP | -0.12 | -4.11 | I(1) | 0.08 | 0.05 | I(0) |
| Inflation | -2.94 | -7.87 | I(0) | 0.41 | 0.03 | I(1) |
| Interest rate | -0.83 | -5.08 | I(1) | 0.11 | 0.09 | I(0) |
| Debt | -1.67 | -4.48 | I(1) | 0.12 | 0.13 | I(0) |
Stationarity Tests and Order of Integration of the Variables
| ADF test | KPSS | |||||
|---|---|---|---|---|---|---|
| Level | First diff. | Order integ. | Level | First diff. | Order integ. | |
| Real fiscal expenditures | -1.92 | -4.09 | I(1) | 0.26 | 0.06 | I(1) |
| Real fiscal revenues | -1.73 | -5.99 | I(1) | 0.10 | 0.06 | I(0) |
| Real GDP | -0.12 | -4.11 | I(1) | 0.08 | 0.05 | I(0) |
| Inflation | -2.94 | -7.87 | I(0) | 0.41 | 0.03 | I(1) |
| Interest rate | -0.83 | -5.08 | I(1) | 0.11 | 0.09 | I(0) |
| Debt | -1.67 | -4.48 | I(1) | 0.12 | 0.13 | I(0) |
Variables of Interest
Citation: IMF Staff Country Reports 2009, 233; 10.5089/9781451813739.002.A002
Source: Fund staff estimates.Variables of Interest
Citation: IMF Staff Country Reports 2009, 233; 10.5089/9781451813739.002.A002
Source: Fund staff estimates.Variables of Interest
Citation: IMF Staff Country Reports 2009, 233; 10.5089/9781451813739.002.A002
Source: Fund staff estimates.
Structural Revenue and Expenditure Shocks
Citation: IMF Staff Country Reports 2009, 233; 10.5089/9781451813739.002.A002
Source: Fund staff estimates ; historical fiscal changes based on Biau-Girard (2005).Structural Revenue and Expenditure Shocks
Citation: IMF Staff Country Reports 2009, 233; 10.5089/9781451813739.002.A002
Source: Fund staff estimates ; historical fiscal changes based on Biau-Girard (2005).Structural Revenue and Expenditure Shocks
Citation: IMF Staff Country Reports 2009, 233; 10.5089/9781451813739.002.A002
Source: Fund staff estimates ; historical fiscal changes based on Biau-Girard (2005).
Impulse Response Functions in the Model Without Debt
Citation: IMF Staff Country Reports 2009, 233; 10.5089/9781451813739.002.A002
Source: Fund staff estimates.Impulse Response Functions in the Model Without Debt
Citation: IMF Staff Country Reports 2009, 233; 10.5089/9781451813739.002.A002
Source: Fund staff estimates.Impulse Response Functions in the Model Without Debt
Citation: IMF Staff Country Reports 2009, 233; 10.5089/9781451813739.002.A002
Source: Fund staff estimates.
Impulse Response Functions in the Model With Debt
Citation: IMF Staff Country Reports 2009, 233; 10.5089/9781451813739.002.A002
Source: Fund staff estimates.NB: Dotted lines represent the results in the baseline model, the solid line the results in the model with debt feedback.Impulse Response Functions in the Model With Debt
Citation: IMF Staff Country Reports 2009, 233; 10.5089/9781451813739.002.A002
Source: Fund staff estimates.NB: Dotted lines represent the results in the baseline model, the solid line the results in the model with debt feedback.Impulse Response Functions in the Model With Debt
Citation: IMF Staff Country Reports 2009, 233; 10.5089/9781451813739.002.A002
Source: Fund staff estimates.NB: Dotted lines represent the results in the baseline model, the solid line the results in the model with debt feedback.Prepared by Brieuc Monfort.
The third chapter of this Selected Issues Papers presents an illustration of the insights derived from one such model, the GIMF model used at the IMF (Yakadina and Yontcheva, (2009).
Interestingly, the increment in debt, rather than the level itself, matters for the VAR, a result similar to the one obtained by Dai and Philippon (2005) in different setup with a yield curve added to the traditional fiscal VAR. They interpret this as meaning that today’s deficit is the best proxy for tomorrow’s debt.
The interpretation of the interest rate in the Favero-Giavazzi setup is thus slightly different than in the Perotti setup: in the latter, the interest rate captures principally monetary policy shocks, as the variable of interest is a short-term interest rate; by contrast, in the former, the implicit interest rate on debt is a weighted average of short and long term interest rates, providing a less clear-cut interpretation. Dai and Phillipon (2005) provides another framework with debt feedback on fiscal variables, with an explicit discussion of the impact on the yield curve through the introduction in their estimation of both a short- and long-term interest rates.
For simplicity, the text and the in-text table refer to the multiplier impact for an expansionary fiscal shock (expenditure increase but tax decrease). The expected sign for the Keynesian multiplier is thus positive, while a negative multiplier indicates that possible Ricardian effects dominate. By contrast, all the charts show the impact of a one-off positive shock of each variable (increase of GDP by one percentage point, increase of fiscal revenues or expenditures by 1 percentage point, increase of inflation or interest rate by 100 basis points).
More specific recommendations on fiscal consolidation—on which the SVAR results could offer no insights—are also worth mentioning. For example, Kumar, Leigh, and Plekhanov (2007) also find that a number of episodes of consolidation were accompanied by policy coordination at different tiers of government; structural reforms, including the introduction of medium-term framework; and reforms of health and pension benefits.
