Model specification

In this chapter, a modification of the simple fiscal rule presented as the model (3.1)

above is considered. Firstly, my model will capture time-varying fiscal responses by relaxing the restriction (time invariability) put on the estimated debt response parameters, and allowing a fiscal rule to change over time. Secondly, the traditional

approach based on a linear response rule in the spirit of (Bohn,1998,2008) allowing

for persistence of fiscal measures, is complemented with an explicit set of controls:

(3.3) bst=ct+%tbst−1+αtdt−1+ΓtΘ(.)+ξt,

wherebst, bst−1 represent the current and lagged primary balance respectively, dt−1

is the one period lagged debt (all variables in percent of GDP),43 Θ(.) stands for a

set of determinants (economic, financial, ...) of fiscal policy exactly specified below

42

Regimes are usually found to be rather persistent with low switching probabilities, which may or may not be a good description of fiscal policy. In the case of imperfect information, changes of regimes may remain unidentified. Similarly, the existence of temporary sub-regimes may not be allowed for.

43_{I employ a model with the one period lagged debt variable followingGreiner et al.(2007) and}

CHAPTER 3. INSTITUTIONAL CHANGES AND FISCAL POLICY BEHAVIOUR

and random shocks (ξt), and ct is the constant term. %t, αt, and Γt represent a

set of time-varying coefficient to be estimated.44 Regarding the determinants, the

literature has yet to reach a consensus as to which are the most important (apart from the debt variable and a proxy for economic conditions). As a result, I consider two versions of the fiscal rule:

(i) Fiscal Rule I(FR I, henceforth) – a model that resembles a Leeper(1991)’s fiscal rule, combining the very simple idea of fiscal policy responding to debt, while controlling for business cycle related fluctuations. This rule is close to the (macro) sustainability strand of the literature,

(ii) Fiscal Rule II (FR II, henceforth) – a specification in accordance with the

model outlined inKirsanova et al.(2005) and briefly outlined above (‘optimal

fiscal policy’), with an inflation rate and a proxy for taxation motives in line

with the original Barro (1979)’s argument (and thus more in the vein of the

FTPL literature).

Models such as the reduced-form fiscal rule (3.1) are meant to capture short-

run fiscal policy behaviour at a yearly frequency, not explicitly considering lags related to policy-makers actions. Since policy-making reality is different, one can specify a fiscal rule that explicitly accounts for these lags. This can also be viewed as an example of the differences between fiscal and monetary policy, as reflected in their policy rules. In addition, since my model will be estimated at a quarterly frequency, a simple transposition of a yearly-based model to the quarterly environment may not be intuitive. Given the typical set-up of fiscal policy, fiscal responses could possibly be more aligned in a four-quarter-change period (that is one year), than those from quarter to quarter, as is commonly done in the literature by bringing

a Bohn’s yearly model to quarterly data.45 However, unexpected events requiring

immediate fiscal intervention (such as catastrophes) often result in ad hoc quarter-to-

quarter changes [(q+ 1)_{/(q)] of government expenditures (structural measures). These}

are rather limited in terms of their size during a fiscal year. Similarly, while (big) changes in taxation are only very rarely carried out on a quarterly basis because of their implications for the behaviour of economic subjects, tax rates do experience changes during a year. In addition, government’s plans for a fiscal year reflect not only quarterly, but also monthly variation, based on the matching of revenue and

44

Based on estimated values, long-run (steady state) levels of debt for the analysed country can be calculated. If_bct<0, then debt in such an economy will level out to a positive long-run level of

debt ratio.

45_{Only few studies have considered such differences so far, for exampleBurger and Marinkov}

CHAPTER 3. INSTITUTIONAL CHANGES AND FISCAL POLICY BEHAVIOUR

expenditure streams. Given the system of checks on budgetary development by the European Commission, governments that clearly deviate from planned (agreed)

paths are expected to correct them without delay.46 Therefore, the assumption

of fiscal policy acting on a quarterly frequency seems plausible. Moreover, one

additional exception can be the case of crisis management to stabilise fiscal accounts (public finances) in fiscally distressed countries, for example the measures recently adopted in GIIPS countries such as Greece and Ireland. Most of the fluctuation can be thus attributed to the role of automatic fiscal stabilisers and planned discretionary measures (such as investments).

In this chapter I begin with a model including only one lag for the sake of comparability with the literature. Further alternatives such as time-invariant estimates or a possible four-lag specification consistent with yearly frequency, are left as a robustness exercise. Therefore, the FR I specification for quarterly series

based on the macro-fiscal rule (3.3) resembling Leeper, 1991’s model (see above)

includes only a control for business cycles and takes the form:

(3.4) bst=ct+%tbst−1+αtdt−1+ΓtΘ0_(.)+ξt,

whereΘ0_(.)∈ {ogapt−1}, where the output gap series (ogap) is estimated using the

Baxter-King band-pass filter (or for robustness check using the HP filter, see data

section or data appendix for details) expressed in relative terms andξt is the error

term. Interpretation of remaining variables is the same as above.

The alternative specification (resembling ‘optimal fiscal rules’, see above) with inflation rate and expenditure/tax smoothing variables (FR II) takes the form:

(3.5) bst=ct+%0tbst−1+α0tdt−1+Γ0tΘ00(.)+ξ

0 t,

where Θ00_(.) ∈ {ogapt−1, πt−1, gcoutt−1}, with πt−1 is the rate of inflation (CPI, in

percent) andgcoutt−1 represents the trend deviations of total current expenditures

(estimated using the Baxter-King band-pass filter or for robustness check using the

HP filter, see appendix for details), andξ_t0 stands for the error term. Interpretation

of remaining variables is the same as above. Based on theoretical considerations

and the work of (Bohn, 1998, 2008), the coefficient %t on lagged primary balance

is expected to be positive (persistence of fiscal measures). The coefficient for the

46

This has become even more important since 2011 when the new procedure for correcting imbalances (MIP) came into force, complementing the already existing ‘European Semester’ (since 2010), see for exampleB´enassy-Qu´er´e(2016).

CHAPTER 3. INSTITUTIONAL CHANGES AND FISCAL POLICY BEHAVIOUR

variable of main interest, the lagged debt ratio, αt would be positive as long as

the government responds with a reduction of primary balance deficit (that is, an

increase of primary surplus) for rising debt andvice versa.

Debt response coefficients estimated in both models (3.4) or (3.5) represent

short-run fiscal responses (α_bSR_t ≡ α_bt), which can be viewed only as partial infor-

mation. Therefore, estimates of long-run fiscal responses (α_bLR

t ) can be added, and

calculated for a particular time periodt, following the logic for ARDL models out-

lined inChud´ık et al.(2015) as:

(3.6) α_bLR_t = αbt

1−%_bt

where %_bt is the estimated coefficient on the lagged primary balance variable from

the model (3.3) for period t.

Some studies have flagged/raised the issue of estimating fiscal responses with

the lagged dependent variable – the models (3.4) or (3.5) – because of the effects

of automatic stabilisers for fiscal responses.47 _{Even though this particular problem}

seems to be relevant for reduced versions of fiscal rules, not explicitly controlling for other determinants but output gap (and for cyclically adjusted primary balance

as the dependent variable), which is not the case here. Since such a model is

(dynamically) misspecified, an appropriate estimation technique is necessary, see

for example Davidson and MacKinnon (2004).48 In addition, such a specification

(3.7) does not allow to distinguish the time frame of the response as with the case

of a model with the lagged dependent variable:

bst=ct+α0tdt−1+ΓtΘ0_(.)+κt, (in case of the FR I)

bst=ct+α0tdt−1+Γ0tΘ00(.)+κt, (in case of the FR II)

κt=גκt−1+ξt,

(3.7)

whereΓtΘ00_(.) orΓt0Θ00(.) includes the same variables as in (3.4) or (3.5) respectively,

κtis theAR(1) error term,גis the estimate of the autoregressive coefficient for the

AR(1) process and ξt is assumed to be aniiderror term.

47_{For example seeGolinelli and Momigliano(2009);Fourier and Fall(2015) orPl¨odt and Reicher}

(2015).

48_{An estimation of the FR I and/or the FR II without the lagged term with anAR(1) error term}

CHAPTER 3. INSTITUTIONAL CHANGES AND FISCAL POLICY BEHAVIOUR