Flexible Fiscal Rules and Countercyclical Fiscal Policy1
  • 1 0000000404811396https://isni.org/isni/0000000404811396International Monetary Fund

Contributor Notes

This paper assesses the impact of different types of flexible fiscal rules on the procyclicality of fiscal policy with propensity scores-matching techniques, thus mitigating traditional self-selection problems. It finds that not all fiscal rules have the same impact: the design matters. Specifically, investment-friendly rules reduce the procyclicality of both overall and investment spending. The effect appears stronger in bad times and when the rule is enacted at the national level. The introduction of escape clauses in fiscal rules does not seem to affect the cyclical stance of public spending. The inclusion of cyclical adjustment features in spending rules yields broadly similar results. The results are mixed for cyclically-adjusted budget balance rules: enacting the latter is associated with countercyclical movements in overall spending, but with procyclical changes in investment spending. Structural factors, such as past debt, the level of development, the volatility of terms of trade, natural resources endowment, government stability, and the legal enforcement and monitoring arrangements backing the rule also influence the link between fiscal rules and countercyclicality. The results are robust to a wide set of alternative specifications.

Abstract

This paper assesses the impact of different types of flexible fiscal rules on the procyclicality of fiscal policy with propensity scores-matching techniques, thus mitigating traditional self-selection problems. It finds that not all fiscal rules have the same impact: the design matters. Specifically, investment-friendly rules reduce the procyclicality of both overall and investment spending. The effect appears stronger in bad times and when the rule is enacted at the national level. The introduction of escape clauses in fiscal rules does not seem to affect the cyclical stance of public spending. The inclusion of cyclical adjustment features in spending rules yields broadly similar results. The results are mixed for cyclically-adjusted budget balance rules: enacting the latter is associated with countercyclical movements in overall spending, but with procyclical changes in investment spending. Structural factors, such as past debt, the level of development, the volatility of terms of trade, natural resources endowment, government stability, and the legal enforcement and monitoring arrangements backing the rule also influence the link between fiscal rules and countercyclicality. The results are robust to a wide set of alternative specifications.

I. Introduction

The aftermath of the recent Great Recession has seen renewed calls for the use of fiscal policy as a countercyclical instrument, owing to the large and protracted growth and employment costs of the crisis, the limited power of monetary policy when interest rates are at the zero-lower-bound, and the perceived potential for increased public investment to avoid a “secular stagnation” in this environment. At the same time, calls have also abounded for a more decisive strengthening of fiscal institutions, and particularly of fiscal rules, as an instrument to ensure prudent fiscal management and bring public debt ratios to safer levels. There is a tension between the two recommendations, as fiscal rules have often been associated with a procyclical bias, and activist fiscal policy with a weakening of fiscal discipline.2

Empirical studies have generally found fiscal rules to be discipline-enhancing (Alesina and Bayoumi, 1996; Bohn and Inman, 1996; Brzozowski and Gorzelak, 2010; Debrun and others, 2008; Fatas and Mihov, 2006; de Haan and others, 1999; Hallerberg and von Hagen, 1997; Manasse, 2006; Perotti and Kontopoulos, 2002; and Tapsoba, 2012). However, the evidence regarding their impact on the cyclical stance of fiscal policy is largely inconclusive. On the one hand, a number of papers have concluded that governments subject to fiscal rules are more prone to procyclical fiscal behavior (Alesina and Bayoumi, 1996; Alt and Lowry, 1994; Lane, 2003; Levinson, 1998; Poterba, 1994; Roubini and Sachs, 1989; and Sorensen and others, 2001). Among spending categories, investment outlays have been found to be more procyclical—countries under pressure to reduce their budget deficits find it politically easier to cut public investment than current outlays (Arezki and Ismail, 2013; Blanchard and Giavazzi, 2004; Dessus and others, 2013). However, other empirical work found that numerical fiscal rules had been associated with less procyclical fiscal behaviors in the European Union (Galí and Perotti, 2003 and Manasse, 2006).

More recently, Ayuso-i-Casals and others (2007), Bova and others (2014), and Combes and others (2014) concluded that fiscal rules could be associated with more countercyclical fiscal policy, provided their design allowed for flexibility, including proper escape clauses, cyclically-adjusted targets, or the extension over several years of the timeframe needed for assessing the compliance with the rule. In a close vein, Bergman and Hutchison (2015) pointed out that fiscal rules are very effective in curbing procyclical fiscal policy once a minimum threshold of government efficiency or quality has been reached.

This paper tries to expand our understanding of the links between fiscal rules and the cyclicality of fiscal policy on three counts. First, it differentiates among types of fiscal rules and explores whether more flexible rules are associated with more, or less, cyclicality. Second, it looks in parallel at the cyclicality of overall spending and that of investment spending. This is important because of the potential growth-enhancing properties of public investment, especially during periods of economic slack and when investment efficiency is high (Afonso and Furceri, 2010; Barro, 1990; IMF, 2014; and Lucas, 1988). And third, it uses propensity scores-matching techniques, borrowed from the microeconomic literature on impact analysis, to handle the self-selection issue that arises from the fact that a country’s decision to introduce a fiscal rule may well be correlated with factors that also affect the cyclical stance of its fiscal policy.

We find that not all fiscal rules have the same impact on the cyclicality of fiscal policy: the design of the rule matters.

  • Among standard rules, budget balance rules are associated with countercyclical changes in overall spending and in investment spending. Expenditure rules are associated with countercyclical changes in overall spending, but with procyclical changes in investment spending, as cuts in the latter during bad times are more politically palatable. Debt rules do not appear to affect the cyclical stance of either overall or investment spending.

  • Flexibility in design seems however, to have the strongest impact. Specifically, investment-friendly rules, or those where public investment or other priority outlays are excluded from the perimeter of the rule, are associated with larger countercyclical movements in both overall public spending and investment public spending. The inclusion of cyclical adjustment features in spending rules yields broadly similar results. The adoption of cyclically-adjusted BBRs is associated with countercyclical movements in overall spending, but with procyclical movements in investment spending. The introduction of escape clauses in fiscal rules does not seem to affect the cyclical stance of fiscal policy.

  • We also confirm that structural factors, including past debt-to-GDP ratio, the level of development, the volatility of terms of trade, natural resources endowment, government stability, the legal enforcement, and monitoring arrangements backing the rule, can influence the link between fiscal rules and countercyclicality. The results are robust to a wide set of alternative specifications.

These findings suggest that an expenditure rule, and to a lesser extent a budget balance rule, may cohabit with countercyclical fiscal policy when investment spending or other priority spending is excluded from the rule target. These findings are in line with recent studies, which concludes that the introduction of investment-friendly rules could help increase investment spending without necessarily undermining fiscal discipline and public debt sustainability, should investment efficiency be high (Blanchard and Giavazzi, 2004; IMF, 2014; IMF 2015b). However, the larger countercyclicality of fiscal policy found in the present paper to be associated with investment-friendly rules is not synonymous with superiority of investment-friendly rules compared with other types of rules. Indeed, investment-friendly fiscal rules may give rise to creative accounting practices, as the lack of a clear-cut conceptual distinction between current expenditure and investment expenditure may provide an incentive for opportunistic misclassification of unproductive expenditures as ‘investment,’ with a view to circumventing the binding constraint of the fiscal rule (IMF, 2014; Serven, 2007)

Of particular importance, fiscal rules have traditionally been enacted to counter the deficit bias and foster fiscal discipline, though a large body of literature has emphasized unpleasant side effects, including procyclicality (Blanchard and Giavazzi, 2004) and lower quality spending (Peletier, Dur, and Swank, 1999). This paper aims at assessing whether certain design features of fiscal rules can alleviate those side effects. Our empirical analysis shows that design matters, which carries potentially important implications for the design of fiscal rules in cases where these side effects are believed to be large. The operational challenge of course is to amend rules in a way that does not jeopardize their effectiveness, an issue we plan to take up in future research.

The rest of the paper is structured as follows. Section 2 introduces the dataset and highlights key stylized facts. Section 3 describes the methodological approach. Section 4 discusses the results and their robustness. Section 5 explores whether structural factors could affect the results. Section 6 concludes and draws some policy implications.

II. Data and Stylized Facts

Fiscal rules, or “permanent constraints on fiscal policy, expressed in terms of a summary indicator of fiscal performance” (Kopits and Symansky, 1998), have multiplied over the past decades. A quick glance at Figure 1 illustrates that by the end of 2012, 80 countries had some type of fiscal rule in place, compared to less than a dozen in the early 1990s. Fiscal rules are usually differentiated by the type of fiscal indicator that they target. Budget balance rules are most common, followed by debt rules, expenditure rules, and revenue rules far behind.

Figure 1.
Figure 1.

Fiscal Rules Adoption over Time (worldwide)*

Citation: IMF Working Papers 2016, 008; 10.5089/9781513581460.001.A001

*Note: this includes FR countries not retained in the sample considered in this analysis

Over time, fiscal rules have become increasingly flexible in their design (Budina and others, 2012). Investment-friendly rules, which exclude public capital spending from the constraint, are the oldest form of flexible rule: they were adopted by some advanced and developing economies as early as the 1970s and 1980s. Investment-friendly rules seek to give space for potentially growth-enhancing public investment while maintaining fiscal discipline. Investment-friendly rules have been criticized for justifying fiscal laxity and encouraging opaque “creative” accounting, but have attracted renewed interest in the 2000s, as evidence emerged that standard fiscal rules were often associated with sizable cuts in public investment and the emergence of “infrastructure gaps” (Servén, 2007; Blanchard and Giavazzi, 2004).

Other approaches have been explored to introduce flexibility within a fiscal rule. Some rules exclude other specific types of spending, such as social transfers or interest payments, from the constraint. Some rules define their targets in cyclically adjusted or structural terms, to allow flexibility to respond to the cycle. More recently, a growing number of fiscal rules have come to include escape clauses that allow for temporary deviations in the case of a large, unexpected shock. Overall, by 2012, 63 countries, or close to 80 percent of those using fiscal rules, had incorporated some form of flexibility in their rules. Of these, 45 had rules with escape clauses, 31 had rules that excluded investment or other priority spending, and 14 had rules that defined targets in cyclically adjusted or structural terms.

In practice, there are sizable overlaps between rules, making an assessment of the impact of a specific rule feature particularly challenging. For example, investment-friendly and cyclically-adjusted features are mostly associated with budget balance rules (Figure 2): budget balance rules account for 34 of the 36 cases of investment-friendly rules, and all of the 15 cases of cyclically-adjusted rules. By contrast, investment-friendly and cyclical adjustment features were less often enacted in the presence of spending rules: spending rules account for 14 of the 36 cases of investment-friendly rules and 8 out of the 15 cases of cyclically-adjusted rules.3

Figure 2.
Figure 2.

Overlaps between Standard Rules (BBR and ER) and Flexible Rules (IR and SR)

Citation: IMF Working Papers 2016, 008; 10.5089/9781513581460.001.A001

1/The numbers refer to the sample retained in this study.

A. Dataset and Measure of Cyclicality

To explore the impact of fiscal rules on the procyclicality of fiscal policy, we use a broad, unbalanced panel of 167 countries over the period 1990–2012; the scarcity of reliable fiscal data prior to 1990, especially for developing countries and ex-Soviet Union members, prevents the use of a longer data period. Out of this sample, 82 countries had fiscal rules in place for at least one year between 1990 and 2012 (Table 1).4 Among these “fiscal rule (FR) countries,” 36 countries introduced investment-friendly rules that shield public investment or other priority spending from the perimeter of the rule. The remaining 85 countries in the sample did not adopt any form of fiscal rule throughout the chosen period. To ensure reasonable comparability across groups, the sample of non-FR countries excludes countries with a real per capita GDP lower than that of the poorest FR country, and a smaller population than the smallest FR country.5

Table 1.

Fiscal Rule Countries (number)

article image
Source: IMF.

The sum of categories may be larger than the total as some countries use multiple rules.

The rule targets the budget balance, usually as a percent of GDP.

The rule targets the level of public debt or public borrowing.

The rule targets the level or growth rate of public spending.

The rule targets the level of public revenue.

Public investment is excluded from the target of the rule.

Public investment and/or other specified spending categories are excluded from the target of the rule.

The target of the rule is defined in cyclically adjusted or structural terms.

Our data only captures the existence of a rule, but not the actual degree of implementation and observance of the rule, for which comprehensive, homogenous data is unfortunately not available. Information on the features of fiscal rules and the dates they were in place come from the 2013 vintage of the IMF Fiscal Affairs Department’s Fiscal Rule Dataset; detailed information on the sample can be found in Appendices 13. Data on total public spending and public investment spending, used to calculate the cyclical stance of fiscal policy, comes from the IMF Fiscal Affairs Department’s expenditure database. Appendix 4 documents the sources and definitions of the variables used in this study. Descriptive statistics are in Appendix 5.

To measure the cyclicality of the fiscal stance, we compute country-specific, time-varying cyclicality coefficients. This approach allows capturing the fact that a government’s reaction to business cycle fluctuations may vary over time or differ between the up and down phases of the cycle. Following Aghion and Marinescu (2008), we estimate the fiscal reaction function (1) with Local Gaussian-Weighted Ordinary Least Squares (LGWOLS):

ΔlogGit=αit+BitΔlogYit+ɛit(1)

with ɛitN(0,σ2ωt(τ)) and ωt(τ)=1σ2πexp((τt)22σ2).

Subscripts i and t refer to country and time dimensions; ΔlogY refers to the growth rate of real GDP;6 and ΔlogG stands for the growth rate of public spending (total spending or investment spending).7 The β^it coefficient captures the cyclical behavior of public spending, which is found to be countercyclical if βit < 0, procyclical if βit > 0, acyclical otherwise. Accordingly, the higher β^it the more procyclical (or less countercyclical) total public or investment spending is.

To ensure an unbiased estimate of β^it, we extend equation (1) in three ways: we include the lagged value of the dependent variable, to capture the inertia in public spending; we run equation (1) with 2SLS to address possible reverse causality between changes in public spending and in real GDP; and we add a vector of covariates (X) to mitigate omission bias. As a result, (1) becomes

ΔlogGit=αit+δitΔlogGit1+BitΔlogYit+γitXit+ɛit(2)

Specifically, the change in real GDP is instrumented with its lagged values, and vector X includes the lagged debt-to-GDP ratio, government stability, volatility of terms of trade, trade openness and financial openness, and inflation rate.

Stylized facts

Figure 3 suggests that total public spending was countercyclical on average during 1990–2012 in FR countries as well as in non-FR countries (the coefficient is negative in both cases). However, the degree of countercyclicality was much more pronounced for FR countries. In contrast, investment spending was procyclical in both FR countries and non-FR countries, but more procyclical in the former. This is in line with the findings of a large body of literature that showed public investment spending to be largely procyclical: it expands during booms but falls during slumps (Arezki and Ismail, 2013; Blanchard and Giavazzi, 2004; Dessus and others, 2013).

Figure 3.
Figure 3.

FRs and Procyclicality of Public Spending (1990–2012)

Citation: IMF Working Papers 2016, 008; 10.5089/9781513581460.001.A001

Figure 4 shows the level of cyclicality coefficients among FR countries before and after the adoption of a fiscal rule. It suggests that the adoption of a fiscal rule was associated with a subsequent strengthening of the countercyclicality of public spending, and a reduction in the procyclicality of investment spending.

Figure 4.
Figure 4.

Procyclicality of Public Spending in FR countries (1990–2012)

Citation: IMF Working Papers 2016, 008; 10.5089/9781513581460.001.A001

As illustrated in Figures 5 to 7, the results vary with the type of rules adopted. The changes in the coefficients go in the same direction, but are quite more marked, after the adoption of an investment-friendly rule (Figure 5). But for cyclically-adjusted balance rules (CAR) or well-designed escape clauses (CR), different patterns emerge: the adoption of a cyclically-adjusted balance rule was associated with a subsequent reduction in the countercyclicality of public spending as a whole and a reduction in the procyclicality of investment spending, while the adoption of rules with escape clauses was associated with a reduction in the countercyclicality of overall spending and a strengthening in the procyclicality of the investment spending. By and large, the adoption of fiscal rules seems to reduce the procyclicality of public spending as well as that of investment spending, but investment-friendly rules seem to be associated with a stronger impact.

Figure 5.
Figure 5.

Public Spending Procyclicality in IR Countries (1990–2012)

Citation: IMF Working Papers 2016, 008; 10.5089/9781513581460.001.A001

Figure 6.
Figure 6.

Public Spending Procyclicality in CAR Countries (1990–2012)

Citation: IMF Working Papers 2016, 008; 10.5089/9781513581460.001.A001

Figure 7.
Figure 7.

Public Spending Procyclicality in Escape Clause Rule Countries (1990–2012)

Citation: IMF Working Papers 2016, 008; 10.5089/9781513581460.001.A001

However, these stylized facts only show simple correlations, and do not address possible self-selection problems: if fiscal rules, and more specifically flexible rules, are only adopted by countries with strong fiscal positions, and thus with the capacity to undertake countercyclical policies, the results are biased.

III. Methodological Approaches

We use propensity score matching (PSM), a method borrowed from the impact analysis literature, to address possible self-selection issues.8 PSM consists of pairing countries that adopted a given policy measure (in our case, a fiscal rule) with countries that have not done so, but share certain characteristics associated with both the adoption of the policy measure and the outcome of interest (in our case, the cyclicality of the fiscal policy stance). These characteristics are synthesized in a propensity score that reflects the estimated probability for a country to adopt the given policy measure, conditional upon the defined characteristics. The propensity score is used to identify a control group (of countries not having adopted fiscal rules) that serves as counterfactual to the treatment group (of countries having adopted fiscal rules). Assuming that the variables used to measure the outcome of interest (here, the cyclicality of the fiscal policy stance) are statistically independent of the policy decision (establishment of a fiscal rule), given common characteristics between the treatment group and the control group, then the difference in outcome between the two groups (known in the literature as average treatment effect on the treated, or ATT) can be attributed to the presence of the fiscal rule.9 More specifically, in this study, the average difference in the cyclicality coefficient (as defined above) between the matched FR countries and the non-FR countries, appropriately weighted by the propensity score distribution of the sample, will be used to estimate the causal effect of fiscal rules on the cyclical stance of fiscal policy.

The ATT can be expressed as follows:

ATT=E[(βi1βi0)|FRi=1]=E[βi1|FRi=1]E[βi0|FRi=1],(3)

Where FRi stands for a binary variable equaling 1 if country i has a fiscal rule in place, and 0 otherwise. βi1 | FRi = 1 captures the procyclical behavior of fiscal policy if country i has adopted a fiscal rule, βi0 | FRi = 1 measures the fiscal policy procyclicality that would have been observed should country i had not introduced a fiscal rule. Equation (3) therefore compares the outcome value observed in the treatment group (FR countries) with the outcome value that would have been observed in the same countries should they had not adopted a FR.

With the propensity score (PS) expressed as P(Xi) = E[FRi | Xi] = Pr(FRi = 1 | Xi), where X is a vector of observable variables associated with the decision to adopt a fiscal rule, and P(Xi) < 1 (so that there are comparable control countries, or non-FR countries, for each treated country, or FR country), equation (3) can be rewritten as:

ATT=E[βi1|FRi=1,p(Xi)]E[βi0|FRi=0,p(Xi)](4)

A. Propensity Scores

We estimate the propensity scores with a probit model, and a dummy for a given fiscal rule as the dependent variable.10 We use different dummies to capture the distinct impact of different fiscal rules: FR for any type of fiscal rule; BBR for budget balance rules; DR for debt rules; ER for expenditure rules; IR for investment-friendly rules (whereby public investment and priority sector spending are explicitly shielded from the target under the rule); CAR when the target of the rule is specified in cyclically-adjusted or structural terms; and CR if the rule includes clearly defined escape clauses.11 Because of the overlap between different categories of rules, we also intersect some of these dummies (e.g., ER * IR, CAR * BBR) when relevant.

To ensure robust results, we use seven different algorithms for country matching, in line with the existing literature (Tapsoba, 2012): the nearest-neighbor matching with replacement, which matches each treated country to the n control countries having the closest PS (we consider n = 1 and n = 3); the radius matching, which matches a FR country to the FR countries with PS falling within a radius (or caliper) of length r (we consider a wide radius r = 0.05, a medium radius r = 0.03 and a narrow radius r = 0.01); the regression-adjusted local linear matching, which consists of pairing covariates-adjusted outcomes for the treatment group with the corresponding covariates-adjusted outcomes for the control group using local linear regression weights (Heckman and others, 1998); and (Epanechnikov) kernel matching, which matches a treated country to all control countries weighted proportionately by their closeness (in terms of PS) to the treated country. Since the matching estimator has no analytical variance, we compute standard errors by bootstrapping, in line with Dehejia and Wahba (2002).

We also use two diagnostic tools to check the validity of the conditional independence assumption, and thus of the matching.

  • First, we follow Rosenbaum and Rubin (1985) and report key statistics to assess the balancing properties of the matched versus unmatched observations. For the conditional independence assumption to hold (i.e., no evidence of significant differences between the FR countries’ and non-FR countries’ observable characteristics within the matched data), the standardized bias score and the p-value associated with its t-test statistic have to stand below the 5% rule of thumb and above the critical threshold of 10%, respectively, (see Caliendo and Kopeinig, 2008; Lechner, 1999; or Sianesi, 2004).

  • Second, we use Rosenbaum (2002) bounding sensitivity tests to check whether unobserved heterogeneity could pollute the results: the ATTs could be biased if countries that appear similar in terms of observed covariates actually differ in terms of important unmeasured covariates that influence both the procyclical behavior of fiscal policy and the decision to introduce a fiscal rule. The bounding sensitivity tests identify the size over which unobserved heterogeneity could impair the results (see Appendix 6 for a detailed presentation of the methodology).

B. Control Variables and Robustness Checks

We use a range of control variables to account for macroeconomic and politico-institutional factors associated in the literature with the adoption of fiscal rules and the cyclicality of fiscal policy. As a reminder, the PS estimation does not aim at finding the best statistical model for explaining the probability of FR adoption, but to control, to the extent possible, for variables that could influence both FR adoption and the outcome variable (fiscal policy procyclicality).12 The selection of variables included in the probit model follows closely this central principle. As macroeconomic indicators, we include the past debt-to-GDP ratio, the rate and volatility of economic growth, and the rate of inflation. As political factors also play a pivotal role in the cyclicality of fiscal policy, we include indicators of political stability and the degree of democracy. Finally, on the institutional front, we include the type of presidential regime, the use of majority electoral rules, federal status and participation to a currency union. Appendix 7 provides details on the empirical literature and expected signs for the control variables.

As robustness checks, we augment the probit model with additional macroeconomic and institutional variables, including the squared value of past public debt (in view of a possible non-linearity in the influence of the debt dynamics); the fiscal balance; trade openness; financial openness; the level of development (seized by per capita real GDP); natural resources endowment; institutional quality (proxied by the quality of bureaucracy); the ruling party’s ideology; the degree of government polarization; the size of the population; the dependency ratio (captured by the share of the population aged 65 and above) ;13 the presence of an IMF program; and a dummy for the occurrence of a crisis.

IV. Fiscal Rules and Procyclicality of Public Spending

A. Propensity Scores

Table 2 displays the probit estimates of propensity scores for different fiscal rules. In column 1, wherein the existence of any fiscal rule is the dependent variable, most coefficients are significant and bear the expected signs: the lagged debt-to-GDP ratio, growth instability, inflation, presidential-type regime, and majoritarian election rules are found to affect negatively and significantly the probability of adopting a fiscal rule, while stronger growth performance, political stability, democracy, federal states, and currency union membership enhance significantly the likelihood of joining the club of FR countries. Results remain broadly similar when budget balance rules, debt rules, expenditure rules, investment-friendly rules, cyclically-adjusted balance, and rules with well-established escape clauses, respectively, are the dependent variables (columns 2 to 7).

Table 2.

Probit Estimates of the Propensity Scores

article image
Note: In brackets the robust standard errors. *, **, and *** indicate the significance level of 10%, 5%, and 1%, respectively. Constant terms are include but not reported.

Matching results

Table 3 reports the matching results associated with the propensity scores from the presence of any type of fiscal rule (estimates from column 1 of table 2 above). Regarding the cyclical behavior of total public spending, the coefficients are negative (indicating countercyclicality) but small, and significant in five cases (out of seven pairing methods). When looking at the impact on public investment spending (bottom panel of Table 3), the coefficients are also negative and larger, though they are significant only in four out of the seven matching cases. This would suggest that the introduction of a fiscal rule is not associated with a clear-cut reduction in the pro-cyclicality of fiscal policies.

Table 3.

Matching Results: All Fiscal Rules

article image
Note: Bootstrapped standard errors (with 500 replications) in brackets. *, **, and *** indicate the significance level of 10%, 5%, and 1%, respectively.

Results could vary, however, according to the type of fiscal rule. Table 4 reports the estimated ATTs for BBRs, DRs, and ERs.

  • For BBRs, the results clearly suggest that the adoption of rules is conducive to more countercyclicality for both overall spending and investment spending: coefficients are negative (indicating countercyclicality) and mostly significant (in six out the seven matching techniques), and larger for the latter.

  • DRs appear unrelated to the cyclical behavior of fiscal policy: all the estimated coefficients turned statistically insignificant for overall spending as well as for investment spending.

  • Regarding ERs, coefficients are negative and significant (in six out seven pairing techniques) for overall spending, but positive and significant (in five out seven pairing techniques) for investment spending. The finding that ERs are associated with countercyclical changes in total public spending is in line with the conclusions of previous studies that showed that ERs help curb pressures for additional spending in the presence of budgetary windfalls (Ayuso-i-Casals and others, 2007; and European Commission, 2006). But the procyclical behavior of investment spending associated with ER suggests that when investment outlays are not specifically shielded, they are more likely than other spending to be cut in downturns (and expanded in booms)—a finding also in line with the literature.

Table 4.

Matching Results: BBRs, DRs, and ERs

article image
Note: Bootstrapped standard errors (with 500 replications) in brackets. *, **, and *** indicate the significance level of 10%, 5%, and 1%, respectively.

Table 5 shows the results for investment-friendly rules. The coefficients are negative, significant and larger both for public spending as a whole and especially for investment spending. Among the whole set of rules, investment-friendly rules are the ones thus associated with the strongest and broadest countercyclicality. The results broadly hold when using a narrower definition of IR countries (see Appendix 8 for details).

Table 5.

Matching Results: Investment-friendly Rules (IRs)

article image
Note: bootstrapped standard errors (with 500 replications) In brackets the. *, **, and *** indicate the significance level of 10%, 5%, and 1%, respectively.

To further probe this finding, we run the same exercise excluding investment-friendly rule countries from the treatment group (Table 6), and see a spectacular reversal in the results: most coefficients lose significance, and when they are significant (for public investment spending), they are positive. This suggests that the countercyclicality evidenced for fiscal rules as a whole was in fact largely driven by the presence of investment-friendly rules.

Table 6.

Matching Results: Excluding IR Countries from the Treatment Group

article image
Note: Bootstrapped standard errors (with 500 replications) in brackets. *, **, and *** indicate the significance level of 10%, 5%, and 1%, respectively.

The inclusion of an investment-friendly feature in ERs and BBRs seems to increase the scope for countercyclical stances. For example, when IRs and ERs overlap (IR*ER), that is, when the treatment group comprises countries with expenditure rules that also exclude investment or priority spending from the ceilings, the adoption of rules is associated with negative coefficients that are larger than in the case of IRs or ERs alone—particularly for overall spending (Table 7). When IRs and BBRs overlap (Table 8), the coefficients are also negative, and particularly larger for public investment spending.

Table 7.

Matching Results: IRs and ERs Jointly as Treatment Group

article image
Note: Bootstrapped standard errors (with 500 replications) in brackets. *, **, and *** indicate the significance level of 10%, 5%, and 1%, respectively.
Table 8.

IRs and BBRs Jointly as Treatment Group

article image
Note: Bootstrapped standard errors (with 500 replications) in brackets. *, **, and *** indicate the significance level of 10%, 5% and 1%, respectively.

Table 9 and Table 10 look at the impact of other flexible rules. Cyclically-adjusted balance rules are associated with negative coefficients (indicating countercyclicality) but these are significant only for public spending as a whole, not for investment spending (Table 9). A possible interpretation is that to meet the target, policymakers tend to avoid procyclical adjustments in current outlays by cutting capital outlays, as the latter are not specifically shielded in the design of CAR. In contrast, the results in Table 10 suggest that rules with escape clauses do not protect from a procyclical fiscal stance: the coefficients are negative for overall spending, positive for investment spending, but in neither case are they significant.

Table 9.

Matching Results: Cyclically-adjusted Balance Rules (CARs)

article image
Note: Bootstrapped standard errors (with 500 replications) in brackets. *, **, and *** indicate the significance level of 10%, 5%, and 1%, respectively.
Table 10.

Matching Results: Rules with Escape Clause (CRs)

article image
Note: Bootstrapped standard errors (with 500 replications) in brackets. *, **, and *** indicate the significance level of 10%, 5%, and 1%, respectively.

The inclusion of a cyclical adjustment feature in ERs and BBRs has mixed effects on their countercyclical impact. The combination of ER and CAR gives statistically significant negative coefficients for both overall and investment spending (Table 11); these coefficients are larger than seen for ERs alone, and largely similar to those seen for the combination of ER with IR. The combination of a BBR with CAR yields larger negative coefficients for overall spending, but lower (and barely significant) coefficients for investment spending (Table 12), which seems to confirm the intuition that unless specifically shielded in the rule, investment outlays will be policymakers’ preferred adjustment variable, even when the rule target is defined in cyclically adjusted terms.

Table 11.

Matching Results: CARs and ERs Jointly as Treatment Group

article image
Note: bootstrapped standard errors (with 500 replications) In brackets the. *, **, and *** indicate the significance level of 10%, 5%, and 1%, respectively.
Table 12.

Matching Results: CARs and BBRs Jointly as Treatment Group

article image
Note: bootstrapped standard errors (with 500 replications) In brackets the. *, **, and *** indicate the significance level of 10%, 5% and 1%, respectively.

In sum, we find that not all flexible fiscal rules accommodate a countercyclical fiscal stance. Investment-friendly rules, and more broadly rules that exclude some categories of spending from the rule target, are the ones more clearly associated with countercyclical changes in both total and investment public spending.14 The inclusion of cyclical adjustment features in ERs yields similar results. The results are mixed for cyclically-adjusted BBRs: the introduction of the latter is associated with countercyclical movements in overall spending, but with procyclical changes in investment spending. The introduction of escape clauses in FRs does not seem to have any impact on the cyclical stance of public spending. Investment-friendly ERs and BBRs, and cyclically-adjusted ERs, therefore appear as the most effective in taming the procyclical bias in public spending.

B. Robustness Checks

Diagnostic tests, reported at the bottom of Tables 312, confirm the robustness of the above results. The p value associated with the standardized biases is above the critical threshold of 10 percent in the large majority of cases. The cutting points from Rosenbaum sensitivity tests hover between 1.2 and 3, large enough levels compared to the findings in the literature (Rosenbaum, 2002; and Aakvik, 2001).15

Appendix 9 shows the results obtained using a probit model augmented to account for possible covariates of investment-friendly rules.16 The propensity scores (Table A9.1) remain quantitatively and qualitatively similar across columns The matching results using these “augmented” probit estimates (Table A9.2) all have similar sign as, and are close in magnitude to, those obtained from the non-augmented model in Table 5.

V. The Role of Strucural Factors

Structural factors can magnify or mitigate the impact of a fiscal rule on the cyclical stance of fiscal policy. To explore their potential role, we look at possible non-linearity in the ATTs, through a control function regression approach. Building on Lin and Ye (2009) and Tapsoba (2012), we use the following OLS regression:

Cyclit=α+βitIRit+γitPscoreit+φXit+ψ(IRit×Xit)+ui+vt+ɛit(5)

where Cycl.it refers to the procyclicality of total spending (or alternatively investment spending); IRit to the investment-friendly rule dummy variable;17 pscoreit stands for the estimated PS from the baseline probit model and is included as a control function; Xit is a vector of macroeconomic, political and institutional factors that could give rise to heterogeneity in the ATT; ui and Vt refer to country and time fixed effects, respectively, while εit refer to the stochastic disturbance term. ψ, the coefficient of the interactive term between IR and Xit, catches the heterogeneity features of the treatment effect of IRs.

Table 13 and 14 report the results for total spending and investment spending, respectively. In each table, Column 1 shows the results of a simple OLS linking IRs adoption to the procyclicality of total spending (investment spending) while accounting for the estimated pscoreit. The β coefficient catches the mean difference in procyclicality between countries having enacted IRs and those that have not. In both cases, it is negative and significantly different from zero, and the magnitudes are close to the coefficients from the matching exercise in Table 5 above (-0.263 for total spending and -1.182 for investment spending). The following columns show the ψ coefficients of the interactive term between an investment-friendly rule and a given structural factor.18

Table 13.

Heterogeneity of Treatment Effect of IRs on the Procyclicality of Total Public Spending

article image
Note: Bootstrapped standard errors (with 500 replications) in brackets. *, **, and *** indicate the significance level of 10%, 5%, and 1%, respectively. Constant terms, as well as vector X variables in isolation (without interaction with IR) are included but not reported for space purpose.