VIEWS: 0 PAGES: 10 POSTED ON: 6/19/2012
What Drives a Successful Fiscal Consolidation?∗ a Pablo Hern´ndez de Cos Enrique Moral-Benito May 2012 Abstract Fiscal consolidations are currently in the agenda of ﬁscal authorities in many countries. Using Bayesian model averaging to overcome the problem of model uncertainty, we ﬁnd that growth-enhancing policies and cuts in public wages are the most appropriate ingredients for successfully reducing debt levels and budget deﬁcits. JEL Codes: H30, H62. Keywords: Fiscal policy, Fiscal consolidation, Budget Deﬁcit, Bayesian model averaging. 1 Introduction In order to ensure ﬁscal sustainability many governments are currently undertaking ﬁscal consoli- dation eﬀorts. A crucial issue from a policy perspective is how to succeed in terms of deﬁcit and/or debt reductions when a ﬁscal consolidation is carried out. According to the diﬀerent criteria con- sidered in the literature, a ﬁscal consolidation is successful if the reduction in the debt-to-GDP ratio (or the primary budget balance) is suﬃciently large and persistent (see for instance Alesina and Ardagna, 2010; Alesina and Perotti, 1995). The empirical analysis of the main factors driving success of a ﬁscal tightening manoeuvre is controversial because there is no theoretical model of reference. The usual approach in the literature investigating the determinants of success in ﬁscal consolidation exercises is based on a regression of a dummy variable identifying successful consolidations on a set of candidate determi- nants (e.g. Alesina and Ardagna, 1998; Giudice et al., 2007). Due to the lack of clear theoretical guidance, results typically depend on the particular variables included in the regression. In order to empirically overcome this model uncertainty problem, we avoid the model selection challenge and consider Bayesian model averaging —henceforth BMA— techniques.1 ∗ n Contact: Banco de Espa˜a. pablo.hernandez de cos@bde.es and enrique.moral@gmail.com. We would like to thank Cristian Bartolucci, Eric Leeper, and seminar participants at the Spanish Economic Association Meeting a n in M´laga and Banco de Espa˜a for useful comments and suggestions. We also thank Silvia Ardagna for kindly sharing the data. The opinions and analyses are the responsibility of the authors and, therefore, do not necessarily n coincide with those of the Banco de Espa˜a or the Eurosystem. 1 In particular we consider the Bayesian averaging of classical estimates (BACE) approach discussed in Sala- i-Martin et al. (2004) and based on Raftery (1995). See Moral-Benito (2011) for a recent overview of BMA methods. 1 Empirical results indicate that growth-enhancing policies are found to be the only relevant factor that generates successful (in terms of reducing debt-to-GDP ratios) ﬁscal consolidations. If the focus is on the persistence of the primary deﬁcit reductions, cutting public wages seems to be the only crucial factor explaining successful ﬁscal adjustments.2 These ﬁndings indicate that, in broad terms, only two determinants already identiﬁed in the literature (i.e. macroeconomic factors and spending-based compositions) are robust to model uncertainty. 2 Data The data used in this paper are from the OECD Economic Outlook No. 84. The sample includes annual information for 21 OECD countries3 from 1980 to 2007. The ﬁscal variables considered as potential determinants of successful ﬁscal consolidations can be classiﬁed in three diﬀerent groups: (i) those variables related to the ﬁscal situation of the country (e.g. government debt and primary budget balance as a share of GDP); (ii) variables capturing the composition of the ﬁscal consolidation (or stimulus) program (e.g. current primary expenditure, government wage and non-wage expenditures, subsidies, income taxes, social security contributions...); and ﬁnally, (iii) the change in cyclically adjusted primary deﬁcit as a proxy for the size of the ﬁscal manoeuvre. The dataset also incorporates a set of macro variables such as the GDP growth rate, the output gap, the exchange rate, and the short-term interest rate. Additional information on the variables considered can be found in the Appendix. 3 Determinants of Successful Consolidations Despite other criteria have been employed in the literature (see below), we ﬁrst focus on the debt reduction criterion considered in Alesina and Ardagna (2010) among others (i.e. successful ﬁscal adjustments are those in which the cumulative reduction of the debt-to-GDP ratio three years after the beginning of the adjustment is greater than 4.5 percentage points). Understanding under which circumstances a ﬁscal consolidation might succeed in reducing the level of debt (or the primary deﬁcit) is relevant from an empirical point of view. One common approach in the literature is based on constructing a successful consolidation dummy that takes the value one for those country-years in which the consolidation succeeded in terms of the previous deﬁnition, and zero otherwise.4 Then, they run a regression of this dummy on a set of macro and ﬁscal variables capturing the environment in which the successful consolidation took place. Depending 2 The rationale for this result comes from the investment channel described in Alesina et al. (2002) who emphasize that deﬁcit reductions achieved through spending cuts from the wage bill (rather than tax increases) are more likely to be successful. This is so because cutting public wages might generate downward wage pressures in the private sector that result in higher levels of investment. 3 Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Japan, Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, United Kingdom and the United States 4 Successful consolidations are isolated from the pool of ﬁscal consolidations identiﬁed by Alesina and Ardagna (2010). According to their deﬁnition, a ﬁscal consolidation episode takes place in a given year if the cyclically adjusted primary balance (CAPB) improves by at least 1.5 per cent of GDP. 2 on the t-statistics of this regression, they conclude which are the most relevant characteristics surrounding a successful ﬁscal consolidation. In broad terms, researchers aim to disentangle the importance of four competing explanations of successful consolidations proposed in the literature: (i) the country’s ﬁscal situation prior to the consolidation, proxied by, for example, the government debt as a share of GDP in the previous year (e.g. Perotti, 1999); (ii) the size of the adjustment proxied by the change in primary deﬁcit during the episode (e.g. Giavazzi and Pagano, 1996; Giavazzi et al., 2000); (iii) the composition of the adjustment in terms of the change in the diﬀerent items of the public bill as a share of the whole change in the primary deﬁcit (e.g. Alesina and Perotti, 1995; McDermott and Wescott, 1996); (iv) the macroeconomic situation captured through the output gap or the growth rate of GDP (e.g. Lambertini and Tavares, 2003). The papers by Ardagna (2004) — henceforth A04 —, Alesina and Ardagna (1998) — hence- forth AA98 —, and Giudice et al. (2007) — henceforth G07 — are good examples of this approach. Table 1 presents the results we obtain replicating the diﬀerent speciﬁcations considered in these papers.5 [Table 1 here] In view of the results presented in the AA98 column in Table 1, one might conclude that the composition in terms of wage expenditures does not aﬀect the success of the consolidation program. However, according to the A04 column, a consolidation based on cutting public wages is expected to reduce the probability of success in terms of debt reduction.6 On the other hand, the statistical signiﬁcance of the impact of macroeconomic conditions also varies depending on the speciﬁcation considered. Moreover, these estimates might again change if we consider alternative proxies such as other items of the public bill (e.g. non-wage expenditures, transfers, business taxes, income taxes...).7 This situation implies that estimates based on single model approaches are always conditional on the empirical speciﬁcation considered. To overcome these issues we consider BMA methods which represent a promising alternative to the previous approach (i.e. the approach considered in Table 1 based on selecting a single regression and deciding which variable is important depending on its associated t-ratio) when model uncertainty is present. The key idea of BMA is to consider and estimate all the possible regressions given by diﬀerent combinations of regressors, and then report a weighted average as the estimate of interest. Moreover, BMA allows a ranking to be constructed of the variables ordered 5 Note here that some of the results are not exactly replicated because either the sample period or the variables’ deﬁnition is not equal to the original papers. Nevertheless, these results are only an illustration of the model uncertainty problem for the sake of motivation. 6 Ardagna (2004) concludes that stabilizations implemented by cutting public spending lead to higher GDP growth rates, and also that the success of ﬁscal adjustments in reducing debt-to-GDP ratio depends on the size of the contraction and less on its composition. 7 Since the number of potential proxies for the four candidate theories (i.e. ﬁscal situation, consolidation size, consolidation composition, and macroeconomic situation) is enormous, the universe of potential regressions to estimate given all the possible combinations of proxies is very diﬃcult to work with. 3 by their relative importance in the contribution to the model ﬁt, i.e. the Posterior Inclusion Probability (PIP).8 [Table 2 here] Table 2 presents the results when applying BMA to estimate all the candidate regressions in order to investigate which regressors are robust determinants of successful consolidations. Output gap is the only variable for which the posterior inclusion probability (PIP) is higher than the prior inclusion probability,9 whereby the main conclusion emerging from Table 2 is that the output gap is the only robust determinant of successful ﬁscal consolidations in terms of debt reduction. Moreover, its posterior standard error is smaller than its posterior mean. Therefore we can also conclude that the output gap positively aﬀects the probability of success of a ﬁscal consolidation package. This result implies that whatever the composition or the size of the adjustment, the most relevant policies in times of ﬁscal consolidation must be oriented toward the objective of sustained and higher rates of GDP growth. Those consolidations not accompanied by economic reforms aimed at increasing employment and productivity may have more diﬃculties in the reduction of debt-to-GDP ratios.10 3.1 Alternative Deﬁnitions of Successful Consolidations How to deﬁne a successful consolidation is not straightforward, and the literature has considered diﬀerent criteria. In addition to the debt-to-GDP ratio criterion considered in the previous section (which is usually the most common approach in the literature), we also explore the determinants of successful ﬁscal consolidation according to two additional criteria. First, we consider a persistence criterion which identiﬁes as successful those consolidations in which the primary cyclically adjusted budget balance improves by at least three percentage points of GDP over three consecutive years (i.e. between t − 2 and t, between t − 1 and t + 1 or between t and t + 2), and in each year the change in the primary cyclically adjusted budget balance cannot be below −0.5 percentage points of GDP. Second, we also consider an expansionary criterion which indicates that a ﬁscal consolidation succeeds if average trend growth between t and t + 2 is greater than between t − 1 and t − 2 (see Giudice et al., 2007). [Table 3 here] BMA results when considering these alternative deﬁnitions of success are presented in Table 3. Regarding the persistence criterion, only the change in wage expenditures as a share of the total 8 See Raftery (1995) or Sala-i-Martin et al. (2004) for more details. 9 This is a commonly-used criterion for labeling variables as robust / non-robust in the BMA literature. We assume a priori that all variables are equally robust (i.e. prior inclusion probability of 0.5) and we label as robust those variables for which the PIP is higher than 0.5. This would imply that the data support these variables more than the rest of the regressors. Other criteria such as the scale of evidence in Raftery (1995) or the t-ratio interpretation in Masanjala and Papageorgiou (2008) reinforce our reading of the results. 10 Which are the best policies for increasing GDP is a controversial question that is beyond the scope of this paper. 4 change in the primary deﬁcit is a robust determinant of success. This implies that the higher the proportion of the consolidation conducted via reducing public wages, the higher the probability of the adjustment being successful in terms of persistence in the deﬁcit reduction. Cutting public wages is a very costly political decision and, therefore, governments will only reduce the public wage bill when they take seriously the ﬁscal consolidation program, and they are thus more likely to achieve the deﬁcit reduction objective. With respect to the expansionary criterion, given that the prior inclusion probability for each variable is 0.5 and all the PIPs are below this threshold, we conclude that there is no variable robustly aﬀecting the probability of a ﬁscal consolidation being expansionary. In addition, all the variables considered as candidate determinants have posterior standard errors larger than the corresponding posterior means, which reinforces the previous conclusion of no variable robustly correlated with expansionary ﬁscal consolidations. 4 Concluding Remarks In this paper we analyse which factors are the most relevant in generating successful consolidations via model averaging techniques. Successful consolidations are those in which the reduction of the debt-to-GDP ratio three years after the beginning of the adjustment is greater than 4.5%. Our results indicate that, in order to succeed in reducing debt levels, economic growth is the crucial ingredient. Without economic recovery, ﬁscal consolidations may have huge diﬃculties in this respect. On the other hand, we also ﬁnd that cuts in public wages are the key ingredient of ﬁscal consolidations in which persistent reductions in primary budget deﬁcits were achieved. A Data Appendix This section describes the data employed in the paper. All data are from the OECD Economic Outlook Database no. 84. • Government debt level: government gross debt as a share of GDP. • Deﬁcit level: cyclically adjusted primary deﬁcit as a share of GDP (i.e. primary expenses minus total revenue) • Consolidation size: Change in the cyclically adjusted primary balance as a share of GDP. • ∆Wage expenditures: Change in government wage bill expenditures. • ∆Non-wage expenditures: Change in government non wage bill expenditures. • ∆Subsidies: Change in subsidies to ﬁrms. • ∆Transfers: Change in cyclically adjusted transfers as a share of GDP. • ∆Government investment: Change in the gross government consumption on ﬁxed capital. 5 • ∆Income taxes: Change in cyclically adjusted direct taxes on household as a share of GDP. • ∆Business taxes: Change in cyclically adjusted direct taxes on businesses as a share of GDP. • ∆Indirect taxes: Change in cyclically adjusted indirect taxes as a share of GDP. • ∆Other taxes: Change in cyclically adjusted other taxes (diﬀerent from income, business or indirect) as a share of GDP. • ∆S.s. contributions: Change in cyclically adjusted social security contributions paid by employers and employees as a share of GDP. • GDP growth: Yearly growth rate of real per capita GDP for each country. • Output gap: % of potential GDP. • ∆Interest rate: Change in the real short-run interest rates between t + 1 and t. • ∆Exchange rate: Change in the exchange rate between t + 1 and t. Note that all the regressors belonging to the public bill (e.g. ∆Wage expenditures, ∆Subsidies, ∆Indirect taxes,...) are divided by the total change in the primary deﬁcit (∆Item/∆Deﬁcit) to focus on the proportion of the adjustment which was due to a particular item as proxies of the composition. More concretely, an increase in these variables means that a larger share of the change in the primary deﬁcit is due to a change in the particular item of the public bill. For the spending items, an increase in these variables means that a larger share of the increase (reduction) of the primary deﬁcit is obtained by increasing (cutting) the particular spending item. For the revenue items, an increase in these variables means that a larger share of the increase (reduction) of the primary deﬁcit is obtained by cutting (increasing) the particular revenue item of the government budget. References [1] Alesina, Alberto, and Silvia Ardagna (1998) “Tales of Fiscal Adjustment.” Economic Policy, 27: 487-545. [2] Alesina, Alberto, and Silvia Ardagna (2010) “Large Changes in Fiscal Policy: Taxes versus Spending.” Tax Policy and the Economy, Editor: R. Brown, vol. 24. NBER. [3] Alesina, Alberto, and Roberto Perotti (1995) “Fiscal Expansions and Adjustments in OECD countries.” Economic Policy, 21: 205-248. [4] Alesina, Alberto, Silvia Ardagna, Roberto Perotti, and Fabio Schiantarelli (2002) “Fiscal policy, proﬁts, and investment.” American Economic Review, 92: 571-589. [5] Ardagna, Silvia (2004) “Fiscal Stabilizations: When Do They Work and Why.” European Economic Review, 48: 1047-1074. 6 [6] Giavazzi, Francesco, Tullio Jappelli, and Marco Pagano (2000) “Searching for non-linear eﬀects of ﬁscal policy: evidence for industrial and developing countries.” European Economic Review, 44: 1269-1289. [7] Giavazzi, Francesco, and Marco Pagano (1996) “Non-keynesian Eﬀects of Fiscal Policy Changes: International Evidence and the Swedish Experience.” Swedish Economic Policy Review, 3: 67-103. [8] Giudice, Gabriele, Alessandro Turrini, and Jan in’t Veld (2007) “Non-Keynesian Fiscal Ad- justments? A Close Look at Expansionary Fiscal Consolidations in the EU.” Open Economies Review, 18: 613-630. [9] Lambertini, Luisa, and Jose Tavares (2003) “Exchange Rates and Fiscal Adjustments: Evi- dence from the OECD and Implications for EMU.” Boston College Working Paper 576. [10] Masanjala, Winford, and Chris Papageorgiou (2008) “Rough and Lonely Road to Prosperity: a Reexamination of the Sources of Growth in Africa using Bayesian Model Averaging” Journal of Applied Econometrics, vol. 23, pp. 671-682. [11] McDermott, John, and Robert Wescott (1996) “An Empirical Analysis of Fiscal Adjust- ments.” IMF Working Paper No.96/59. [12] Moral-Benito, Enrique (2011) “Model Averaging in Economics.” Bank of Spain Working Paper 1123. [13] Perotti, Roberto (1999) “Fiscal Policy In Good Times And Bad.” The Quarterly Journal of Economics, 114: 1399-1436. [14] Raftery, Adrian (1995) “Bayesian Model Selection in Social Research” Sociological Method- ology, vol. 25, pp. 111-163. [15] Sala-i-Martin, Xavier, Gernot Doppelhofer, and Ronald Miller (2004) “Determinants of Long- Term Growth: A Bayesian Averaging of Classical Estimates (BACE) Approach.” American Economic Review, 94: 813-835. 7 Table 1: Characteristics of Sucessful Fiscal Consolidations AA98 A04 G07 Dependent variable is the successful consolidation dummy GDP growth 12.47 (t-ratio) (4.67) Government debt in t − 1 −0.02 −0.23 (t-ratio) (−0.17) (−1.58) Deﬁcit level in t − 1 0.34 (t-ratio) (0.31) Consolidation Size 1.91 2.52 2.44 (t-ratio) (0.61) (0.90) (0.76) ∆ wage expenditures −0.03 −0.41 0.01 (t-ratio) (−0.18) (−2.07) (0.05) ∆ interest rate −0.01 (t-ratio) (−0.22) ∆ exchange rate 0.01 (t-ratio) (0.32) Output gap 0.02 (t-ratio) (0.82) R2 0.01 0.29 0.06 Obs. 73 73 73 Notes: This Table presents the results from estimating three OLS regressions of the successful (in terms of debt reduction) consolidation dummy on the determinants suggested in Alesina and Ardagna (1998) [AA98], Ardagna (2004) [A04], and Giudice et al. (2007) [G07]. The change in wage expenditures is relative to the change in primary budget deﬁcit (i.e. ∆wage expenditures is one of the items in ∆itemit /∆Defit ; an increase in this variable means that a larger share of the reduction of the primary deﬁcit is obtained by cutting the public wage bill.). 8 Table 2: Characteristics of Successful Consolidations via Model Averaging PIP P. Mean P. Std. Output gap 0.60 0.05 0.02 ∆S.s. contributions 0.42 −0.58 0.30 ∆Other taxes 0.28 −0.57 0.37 ∆Transfers 0.25 0.28 0.21 GDP growth 0.21 4.50 3.64 ∆Business taxes 0.20 0.31 0.27 ∆Government investment 0.16 0.15 0.16 Deﬁcit level 0.15 −1.36 1.61 ∆Non-wage expenditures 0.14 −0.32 0.43 Government debt level 0.13 −0.15 0.18 ∆Wage expenditures 0.13 −0.19 0.31 ∆Indirect taxes 0.12 −0.13 0.26 ∆Interest rate 0.11 −0.02 0.03 ∆Income taxes 0.11 −0.03 0.22 Consolidation size 0.11 −1.87 3.84 ∆Subsidies 0.11 −0.24 0.56 ∆Exchange rate 0.10 −0.01 0.02 Prior Inclusion Probability 0.5 Number of models estimated 131, 072 Notes: PIP refers to the posterior inclusion probability of a particular regressors. Given the prior inclusion probability is equal for all the variables (i.e. 0.5), those variables with PIP higher than 0.5 are labeled as robust determinants of successful consolidations. All the regressors belonging to the public bill (e.g. ∆Subsidies, ∆Indirect taxes,...) are divided by the total change in the primary deﬁcit to focus on the proportion of the adjustment which was due to a particular item as proxies of the composition. P. Mean refers to the posterior mean conditional on inclusion of a given regressor in the empirical model, which is a weighted average of model-speciﬁc coeﬃcient estimates with weights given by the model-speciﬁc R-squares. P. Std. is the square root of the posterior variance which is a weighted average of model-speciﬁc variances also including the variance of the estimates across diﬀerent models. The sample is formed by 73 country-year pairs in which a consolidation took place. The 131, 072 estimated models come from all the possible combinations of the 17 regressors (217 = 131, 072). 9 Table 3: Characteristics of Successful Consolidations: Alternative Deﬁnitions Persistence Criterion Expansionary Criterion PIP P. Mean P. Std. PIP P. Mean P. Std. (1) (2) (3) (4) (5) (6) ∆Wage expenditures 0.72 0.38 0.15 0.10 0.05 0.22 ∆Income taxes 0.34 0.19 0.11 0.11 −0.07 0.16 GDP growth 0.17 2.13 2.00 0.10 −0.51 2.81 ∆Indirect taxes 0.14 −0.12 0.15 0.10 −0.06 0.20 ∆Government investment 0.13 −0.07 0.10 0.13 −0.09 0.13 ∆S.s. contributions 0.13 0.13 0.19 0.16 0.28 0.27 Output gap 0.12 0.01 0.01 0.17 −0.02 0.02 ∆Other taxes 0.12 −0.12 0.25 0.10 −0.10 0.31 ∆Transfers 0.12 −0.06 0.12 0.14 0.14 0.16 ∆Business taxes 0.12 −0.08 0.16 0.11 −0.08 0.22 ∆Non-wage expenditures 0.11 0.12 0.24 0.12 0.25 0.34 Government debt level 0.10 0.04 0.11 0.10 0.04 0.15 Deﬁcit level 0.10 −0.29 0.82 0.19 1.46 1.20 ∆Interest rate 0.10 0.00 0.02 0.10 0.00 0.03 ∆Subsidies 0.10 0.06 0.31 0.10 0.11 0.46 ∆Exchange rate 0.10 0.00 0.01 0.10 0.00 0.01 Consolidation size 0.10 0.11 2.19 0.26 4.84 3.20 Prior Inclusion Probability 0.5 0.5 Number of models estimated 131, 072 131, 072 Notes: PIP refers to the posterior inclusion probability of a particular regressors. Given the prior inclusion probability is equal for all the variables (i.e. 0.5), those variables with PIP higher than 0.5 are labeled as robust determinants of successful consolidations. All the regressors belonging to the public bill (e.g. ∆Subsidies, ∆Indirect taxes,...) are divided by the total change in the primary deﬁcit to focus on the proportion of the adjustment which was due to a particular item as proxies of the composition. P. Mean refers to the posterior mean conditional on inclusion of a given regressor in the empirical model, which is a weighted average of model-speciﬁc coeﬃcient estimates with weights given by the model-speciﬁc R-squares. P. Std. is the square root of the posterior variance which is a weighted average of model-speciﬁc variances also including the variance of the estimates across diﬀerent models. The sample is formed by 73 country-year pairs in which a consolidation took place. The 131, 072 estimated models come from all the possible combinations of the 17 regressors (217 = 131, 072). 10