Fiscal Policymaking in Central and Eastern Europe: What Does 12 Years of Real Time Data Tell Us? John Lewis




старонка1/3
Дата канвертавання19.04.2016
Памер220.5 Kb.
  1   2   3
Fiscal Policymaking in Central and Eastern Europe: What Does 12 Years of Real Time Data Tell Us?
John Lewis

De Nederlandsche Bank

Abstract
This paper estimates fiscal policy reaction functions for CEECs using a real time dataset gathered from successive releases of IMF’s International Financial Statistics Database.
JEL Codes:

E62 (Fiscal Policy),

E61 (Policy Objectives, Policy Designs and Consistency),
Keywords: Central and Eastern Europe, Fiscal Policy, Real Time Data

The dataset used in this paper was partly compiled using old IMF CDs at the Bank of Estonia, who kindly allowed me access to their data archive during a visit to Tallinn in February 2009. I am grateful to the Bank’s research department for providing me with office facilities during this time. Special thanks to Rasmus Kattai for help computing the real time output gap figures. This paper has benefited from the helpful comments of and the empirical analysis was greatly helped by the comments of Steven Poelhekke. The views expressed are those of the author and not necessarily those of De Nederlandsche Bank.



1. Introduction
Over the past five years, a small but growing literature has analysed the determinants of fiscal policy in Central and Eastern European Countries. These papers have typically followed the widespread technique of estimating reaction functions where the fiscal instrument is a function of the output gap, debt, and other variables. Such fiscal policy rules provide a simple description of policymaker’s behaviour and their coefficients have a clear interpretation in terms of the inertia of fiscal policy, responsiveness to cyclical conditions and to other variables.
Recently, there has been increasing emphasis on the use of real time data- i.e. reconstructing the preliminary data available to policymaker at the time of their decision- to analyse monetary and fiscal policymakers’ behaviour.
Up to now, this insight has not been applied fiscal policy in CEECs, probably due to the fact officially published “real time datasets”, do not cover CEECs.1 Using a self-compiled database based on previous releases of IMF data and EBRD transition reports, this paper applies real time data techniques to the analysis of CEEC fiscal policy for the first time. The primary goal of the paper is to examine whether earlier empirical findings for CEECs hold when real time data is used.
In principle, one might expect the difference between real time and ex post data to be greater for CEECs than for more mature market economies. Economic time series are likely to be more volatile, formal econometric modelling is hampered by limited data and frequent structural breaks and the transition process to a market economy was largely unprecedented. On the forecasting side, these factors may make it more difficult to correctly forecast the the level and growth rate of output and to correctly predict the spending and revenue implications of their budgetary plans. On the data collection side, statistical agencies, especially in the early years, may have had less experience and resources in data collection their counterparts in richer countries, and the variables themselves may undergo more frequent methodological changes. These factors suggest that for CEECs, relevant variables for a fiscal reaction function are likely to be more heavily revised between the policymaker’s decision and the econometrician’s regression than in more mature economies.
Orphanides (2001) argues that reaction functions are only a valid account of policymakers intentions if they represent a rule which the policymaker could actually have implemented at the time. That requires that they are expressed in terms of data which was available when the decision was made. This point is likely to be even more relevant in the realm of fiscal policy, because the policy instrument itself may be measured with error, as well as the variables that the policymaker is reacting to. Fiscal policymakers pass a given set of expenditure and tax plans, in the hope of achieving a given deficit/surplus, but the eventual outturn may differ significantly from their own forecasts. The eventual outturn of government expenditure and revenues is a function of both the budgetary measures enacted by the policymaker, and of the state of the economy which can be affected shocks other than those generated by policymakers. As Beetsma and Giuliodori (2008) have emphasised, fiscal policymakers’ plans (as proxied by the real time data) may give a more reliable picture of policymakers intentions than the actual outcome.
Indeed, papers which estimate fiscal policy reaction functions for Western Europe using ex post and real time data have found that data vintage can have an important effect on the results obtained (Cimadomo, 2007; Momigliani and Golinelli, 2006, 2008)- namely that discretionary fiscal policy tends to look acyclical based on ex post data, but counter cyclical when real time data is used. Additionally, those papers which consider the role of measurement errors along side ex post data find that the former are often significant determinants of fiscal policy (Bernoth et al, 2008; Larch and Salto, 2007; Von Kalckreuth and Wolff, 2008).
A number of authors have estimated fiscal policy reaction functions in CEECs, but they have all done so on the basis of ex post data. They find a number of important CEEC-specific findings. Berger et al (2007) estimate pooled fiscal reaction functions for eight CEECs. They find evidence of a significant loosening in larger central European countries after 1999 coincident with these countries accession to NATO. The authors argue that following NATO accession these countries felt assured of EU accession and hence the disciplining effect of potential EU exclusion no longer operated. They found some evidence of electoral cycles affecting fiscal policy, but no discernable influence for institutional variables such as political fragmentation or centralisation. Lewis (2007) finds a similar distinction between the Baltics and the Central European countries- with the former’s fiscal policy typically tighter and more responsive to the cycle than the latter group. Staehr (2007) compares the behaviour of fiscal authorities in “Eastern” and “Western” Europe. Estimating fiscal reaction functions for both groups, he finds that CEEC budget balances are relatively less inert, and more responsive to the cycle than those in Western Europe. Lastly, Fabrizio and Mody (2006) regress budget balances on unemployment (rather than the output gap) and variety of institutional and political variables. However, with the exception of the latter, these studies typically utilise a relatively simple econometric methodology. Lagged dependent variables are omitted, despite their significance in most comparable regressions for Western Europe, and panel data approaches are employed which are typically inefficient in the presence of either weak instruments and persistence in the left hand side variable.
Accordingly, this paper innovates with respect to the literature in three key ways. First, it revisits earlier studies based on ex post data to examine whether their findings are robust to more sophisticated econometric techniques. Second, it represents (to the author’s knowledge) the first attempt to apply real time data techniques to the study of fiscal policy in CEE, and to thus consider whether the choice of data vintage can affect the results.
The paper is organised as follows. Section 2 discusses the dataset used. Empirical results are presented in section 3. Finally section 5 concludes.

2. Data
All data used in this paper is taken from the IMF and the EBRD. Other international bodies did not publish data for all CEECs in real time in their regular statistical publications2, and therefore the EBRD was the only source which provided data for all countries, at every year in the sample period.
To date, neither organisation has released its own real-time database and hence this dataset was constructed by the author. To the author’s knowledge, this dataset is unique in the literature. It covers ten CEECs (Bulgaria, Czech Republic, Estonia, Hungary, Latvia, Lithuania, Poland, Romania, Slovakia and Slovenia) and contains observations for GDP growth, total budget balance and government expenditure at a yearly frequency.3
The EBRD’s transition reports record projected values for the current year and forecasts for the next year as well as earlier observations. This draft of the paper uses all transition reports from 1997 onwards, giving a total of 12 vintages of data. The real GDP growth figures are the same as IMF staff projections.4 This property means the GDP growth figures can be matched up with data on real GDP levels from IMF’s International Financial Statistics (IFS) database.
Table 1: Root Mean Squared Revision in Real Time Data 1997-2007
Table 1 presents the root mean squared revision (RMSR) of the real time data on GDP growth and government budget balance for CEE countries, and comparison groups of Western European countries. Overall the data strongly confirm the conjecture of the introduction that data for CEE countries are typically subject to larger real time measurement errors than their Western European counterparts. For GDP growth, the RMSR between time t and the final data vintage is more than twice the as large in CEE than in Western Europe. Moreover, every individual CEE has a higher RMSR than the Western European average figure. Similarly, for overall budget balance, the RMSR between time t and the final vintage is higher for CEECs. All bar two CEECs have a RMSR above the Western European average.
Also striking is the fact that many countries which have a relatively high RMSR for GDP growth (e.g. Latvia and Romania) have a relatively low RMSR for budget balances. Conversely, some countries whose GDP growth data is typically not revised very much have relatively large revisions to the their budget balance data (e.g. Slovakia, Bulgaria).

The differences across vintages are illustrated in figure 2, which graphs eight separate vintages of GDP growth data for Estonia, from the period 1997-2008.


Figure 2: Eight Vintages of GDP Growth in Estonia
This shows that data revisions often continue many years after the first release. In the case of Estonia, it usually takes about four years worth of revisions before the data “settles down” to a value which is reasonably stable across vintages. Thus, for GDP growth in 1997, it is apparent that all vintages report a reasonably similar figure, reflecting that the fact that the earliest vintage shown (2001) was already four years after the first release. By contrast, for 2002 GDP growth, the figure differs more substantially across vintages. Even four years after the event (i.e. the 2006 data vintage) the figure was revised upwards by almost a percentage point.
3. Empirical Estimates of Fiscal Reaction Functions
The basic form of the fiscal reaction function is as follows:
(1)
where bal is the (total) government budget balance, expressed as a percentage of GDP, growth is real GDP growth and u and v are country and time specific fixed effects. This functional form has been used in other studies of CEE countries (Staehr, 2008, Berger et al 2007) and its use here facilitates comparison with other studies5. In addition, the use of real time data requires that data for fiscal variables, and for the measure of economic activity has to be available in real time. EBRD transition reports do not report primary budget balance data, nor do they attempt any cyclical adjustment of government finances. Thus the total budget balance is used as fiscal variable.6 Similarly, output gaps are not recorded in EBRD transition reports, and their availability (especially in real time) is somewhat patchy for CEECs.
From an economic perspective, it is difficult to believe that policymakers in CEECs (at least in the early part of the sample) could have formulated an accurate real time measure of potential output, and hence of the output gap. Rapid structural change and limited data would have made the construction of such a measure highly troublesome. If policymakers did wish to respond to economic activity, a somewhat cruder measure such as economic growth may have been a more plausible candidate.
3.1 Estimation Methodology
A panel unit root test7 reported no evidence of non-stationarity, therefore estimation proceeds on the basis that the variables in the dataset are stationary. Given the relatively short time period involved, the data is pooled and estimated as a dynamic panel.
To deal with potential simultaneity between economic growth and the government budget balance, the level of economic growth is instrumented. Variables outside the model which are used as instruments are: eurozone economic growth, and the (unweighted) average of the other nine countries economic growth rates. Since each country is relatively small compared to the eurozone (and to its neighbours), it is improbable that one countries fiscal policy could affect economic growth in the CEE region or the eurozone.
Even in the absence of any simultaneity issues, dynamic panels with a lagged dependent variable can lead to inconsistent and biased coefficient estimates even if cross sectional fixed effects are included (see Arellano, 2003). One common resolution is to remove the panel-level effects by first-differencing the regression equation and then applying linear GMM. In this context, the Arellano-Bond estimator is often used, taking lagged levels as instruments for the first difference of the dependent variable. However, Blundell and Bond (1998), showed that when the autoregressive parameter is moderately large (which is often the case in fiscal datasets), then the lagged levels becomes poor instruments for the first differences.
Accordingly, reaction functions are estimated by one-step system GMM (see Blundell and Bond 1998) estimator.8 The second and third lags of the balance and economic growth are used as pre-determined instruments alongside the exogenous instruments listed above.
Alongside the time and country dummies, other simple time based variables were experimented with to attempt to capture possible institutional effects. Berger at al (2007), construct a variable which equal to the number of years since 1999 for the Czech Republic, Hungary, Slovakia and Poland, and zero for all other countries. They find this variable to be significant, and interpret this as evidence of a systematic fiscal loosening in these countries who, having joined NATO, felt the EU accession was a done deal. The first variant BKS1 measures the number of years since 1998 for these four countries. Of course it could be the case that the loosening did not continue for nine years. To account for this possibility, a second variant BKS2 takes the value of five in 2004 and all subsequent years (equivalent to assuming the loosening took five years and then settled at the new level). Alternatively, it may have been that fiscal stance was effected EU accession.9 To test for this, the variable PostEU is constructed measuring the number of years since EU membership (and equal to zero if the country is yet to join).
3.2 Empirical Results: Ex Post Data
The results are shown below in table 3:
Table 3: Fiscal Policy Reaction Functions with Ex Post Data- Time Effects
These results show that a similar story holds across a wide variety of specifications of time and country related effects. The baseline specification (I), finds that fiscal policy does respond countercyclically, but that the response is quite small (a tightening of 0.2% of GDP for every percentage point of growth). Also, fiscal policy does exhibit inertia, but with a coefficient of around 0.210, it is considerably less inert than is seen in Western Europe, where figures of 0.6-0.8 are the commonplace.11 Country effects are always strongly significant whenever included, but the time dummies are never jointly significant. Dropping the time dummies (II), raises the inertia of fiscal policy but makes little difference to the response to growth.
The BKS variable is not significant when it is included in a specification with country effects (III). If country effects are also excluded, (IV), then the BKS variable does become significant, and takes a very similar value to the 0.3 reported by Berger et al (2007). However, this happens only when (otherwise significant) country dummies are suppressed.12
The Post EU variable is not significant when time dummies are included (V). However, the year dummies are not jointly significant, and if they are dropped (VI), the Post EU coefficient is strongly significant, indicating a moderate tightening of fiscal policy after EU accession. If the BKS and PostEU variables are jointly included, the story is similar. With year dummies (VII) neither is significant; dropping the year dummies (VIII) renders the PostEU variable strongly significant.
As a robustness check, the baseline specification (I), was tested against a variety of modifications, the results of which are shown in the appendix. The magnitude and significance of the coefficients on the economic growth and the lagged balance are robust to changes in the outside instruments used, the lag structure of inside instruments and the estimation technique.
To test for possible asymmetries in reaction between good and bad times, a split sample approach was used, where “good” periods are classified by economic growth of more than 5.0%, and “bad” periods where growth is below that figure.13 The results are shown below in table 4.
Table 4: Asymmetric Fiscal Reactions using Ex Post Data
The sample was split up into “good” and “bad” periods on the basis of economic growth, the cutoff point being 4.5 (which splits the sample into two roughly even sub-samples). Running the regression with time dummies included (I) leads to problems concerning over-identification. Since these are not significant in good times, and only weakly significant in bad times, a regression was run without time effects (II). In any event the coefficient estimates are broadly similar across the models, but (II) faces better on the Sargan test. These results demonstrate a clear asymmetry in fiscal policymaking- in good times, fiscal policy tends to be considerably more inert than in bad times. That suggests that fiscal policymakers are more inclined to spend extra revenues as they come in. On the other hand, in bad times fiscal policy gets less inert, and one of the specification indicates an increased responsiveness to economic conditions.
3.3 Empirical Results: Real Time Data
Table 5: Time Effects Using Real Time Data
Estimating the same basic regression with real time data yields different results. In the benchmark model (I), fiscal policy is considerably more inert (the co-efficient doubles to 0.36) then when ex post data is used, and the response to economic growth is slightly stronger. The Sargan test suggests overidentification. Dropping the time variables improves this a little, but with real time data the time effects are always jointly significant wherever included. To remove the problem over identification, the instrument set is collapsed, and orthogonal deviations are used in place of first difference transformation (III – VI). This change leads to significantly better diagnostic results- although the size of the coefficients remains fairly similar. The baseline specification with time dummies yields a higher inertia coefficient, (0.48), but the cyclical response coefficient is almost identical. Dropping the time dummies (IV) leads to a higher inertia coefficient and a slightly lower cyclically adjusted coefficient. Neither the BKS (V) or the PostEU (VI) variable is significant.
Table 6: Testing For Asymmetries with Real Time Data
Table 6 shows the results for the split sample specification between good and bad times. All specifications use collapsed instruments and orthogonal deviations. The results suggest little evidence of an asymmetric reaction with real time data. With time dummies (I) fiscal policy is more inert in bad times, and slightly less response (though neither of the economic growth coefficients are significant). Removing the time dummies (II), gives almost identically size coefficients across good and bad times. Taken with the earlier finding of an asymmetry when ex post data was used, this suggests that the asymmetric reaction may not be a function of deliberate actions by policymakers.
4. Concluding Remarks
This paper estimates fiscal policy reaction functions for Central and Eastern Europe, using both ex post and real time data. On the basis of the ex post data, fiscal policy, as measured by the overall budget balance, appears to react in a countercyclical way over the cycle, and exhibits less considerably inertia than is typically found in Western Europe. There is also a marked asymmetry- in slower growth times, fiscal policy is less inert, and more responsive to output; whereas in times of high growth, budget balances become quite inert, and show little responsiveness to the cycle.
Using real time data shows a slightly stronger cyclical response, and a large increase in the inertia associated with fiscal policy. That suggests that the apparently low inertia seen in CEECs may not reflect the conscious actions of policymakers in the planning and implementation stages. The asymmetry present in the ex post data does not carry over to the real time case- that again suggests that the asymmetry is not necessarily the result of a conscious desire by governments to set fiscal policy in an asymmetric way.
Overall, comparing the real time and ex post results, it is noticeable that the cyclical response of fiscal policy is largely unchanged with respect to the data vintage used. Using real time data tends to raise the cyclical response co-efficient by about eight basis points. By contrast, several studies on Western Europe find that the co-efficient tends to rise by around thirty basis points14. Similarly striking is the fact that for CEE, the inertia coefficient changes substantially, whereas studies in Western Europe report virtually no difference (i.e. less than ten basis points) in the inertia of fiscal policy for alternative data vintages.
The regressions here find little or no role for time effects. The pre-accession fiscal loosening found by Berger et al (2008) only shows up here when both time and country fixed effects are suppressed. When included, these fixed effects indicate that larger central European economies tended to run looser fiscal policies (in all years) and that fiscal policy was looser in the years after 1999 (in all countries). Once these effects are included, the BKS variable has no significance. There is some evidence of a tightening in fiscal policy after EU accession, but this effect is small, and is not significant when real time data is used.
The dataset does not permit a direct decomposition between automatic and discretionary policy, since it contains data only for the unadjusted government budget balance. However, the results can be combined with estimate of budgetary sensitivities to the cycle to gain a rough picture of the relative contributions of automatic and discretionary policy. The European Commission (2008) estimates that automatic stabilisers are around 35 cents in the euro for CEE countries, and is thus larger than estimated budgetary response to economic growth (0.18 to 0.26). That suggests that the discretionary element of fiscal policy may have been working in a pro-cyclical fashion, with governments tending to take expansive measures in good times, and contractionary measures in bad times. This mildly procyclical discretionary policy response is then dominated by the countercyclical effects of automatic stabilisers.

  1   2   3


База данных защищена авторским правом ©shkola.of.by 2016
звярнуцца да адміністрацыі

    Галоўная старонка