DESDE LA VIVENCIA DEL SUJETO Y SU CONTEXTO: LA ELECCIÓN METÓDICA Y METODOLÓGICA DEL ESTUDIO
2.2. Frente al enfoque metódico del estudio.
With the 2005 reform of the Stability and Growth Pact (SGP) the structural budget balance has taken centre stage in the EU fiscal surveillance. All key requirements of the revised Pact are now expressed net of cyclical factors and one-off and other temporary measures (1).
Because of its increased prominence in the assessment of countries' fiscal positions, the structural budget balances is subject to a particularly intense scrutiny both in academic and policy circles. This increased attention has laid bare a number of limitations and methodological issues with tangible policy implications. Magnified by recent developments, two issues have received particular attention: (i) the uncertainty attached to the assessment of cyclical conditions in real time and (ii) significant short-term fluctuations in the tax content of economic growth. A detailed review of the recent experience with the use of the structural budget balance in the EU fiscal framework was included in the 2007 edition of the Public Finances in EMU report.
Following up on this, the present section outlines ways to better cope with the two limitations of the structural budget balances mentioned above. The first subsection takes a closer look on how to reduce the degree of uncertainty attached to real- time estimates of the output gap by making use of additional economic indicators which are less prone to revisions. The second subsection explores a method for tracking and explaining year-on-year changes in the yield of the tax system (2).
(1) This is the case for the Member States' medium-term
budgetary objective (MTO), the annual required budgetary adjustment for countries that have not reached their MTO yet as well as the fiscal adjustment required to correct an excessive deficit. A detailed review of the revised Pact and the role played therein by the structural budget balance can be found in European Commission (2006a).
(2) The options sketched out in both subsections are
currently being discussed and assessed in the competent
calculation of the structural budget balance, after consultation and agreement with the MS, so as to improve its effectiveness in the EU fiscal framework.
2.1.1. Improving the assessment of the output gap in real time
In the EU fiscal surveillance framework, the cyclical component of the budget is obtained by applying an aggregate budgetary sensitivity to estimates of the output gap (see Box II.2.1 for the details of the method). This approach is well established in the literature and puts into practice the understanding that output fluctuations affect the budget balance and may obscure the view on what is generally called the underlying budget balance, i.e. the budget balance that would prevail if output was at its potential level.
The main downside of output gap estimates is their considerable degree of uncertainty. This uncertainty has direct implications regarding the assessment of the cyclically adjusted balance (CAB) in real time . In practice, measures of the output gap available in real time may differ significantly from those constructed with the same method on the basis of data published years later. Such revisions mainly reflect the uncertainty inherent to forecasts: in order to assess the current position in the cycle one inevitably has to make an assumption of where the economy is expected to be in the future. As forecasts are revised, they also affect the current assessment of the cyclical position.
From the viewpoint of fiscal policy making and surveillance, the uncertainty surrounding output gap estimates is a serious issue (3). It can give
rise to a distorted diagnosis and, in turn, to an inappropriate policy response. A relatively recent and particularly evident case in point, reviewed in more detail in the 2007 issue of this report, were the late 1990s, when available data and
working group of the Economic Policy Committee, the Output Gap Working Group.
(3) See Langedijk and Larch (2007) for an assessment of the
sensitivity of the EU fiscal framework to variations in output gap estimates.
prevailing forecasts led to believe that the economies of most euro area countries were operating slightly below potential. With the benefit of hindsight, the output gap in the years 1999-2000 turned out to be abundantly positive and the fiscal stance too lax.
The way forward
A way forward to address the uncertainty attached to real-time output gap estimates is to broaden the assessment of cyclical conditions with a battery of complementary real-time indicators that can reflect cyclical developments. This would be in line with the provisions of the Code of Conduct according to which "the identification of periods of economic 'good' times should be made after an overall economic assessment" (1). The number of potential
candidates for complementary indicators is very large. In principle, all macroeconomic variables and survey indicators that are expected to reflect or mimic cyclical developments can be of use, for instance the rate of inflation, changes in the rate of unemployment, interest rates, real exchange rates, the current account balance or the rate of capacity utilisation. A first attempt to bring on board complementary indicators was made in the 2006/07 assessment round of the stability and convergence programmes. The approach was purely descriptive and judgemental in nature (2).
With a view to making the assessment more systematic two different quantitative methods have been tried out. The first method applies a forecast approach recently developed by Stock and Watson (2002) and makes use of a battery of indicators to forecast the output gap using principal components analysis. This approach is purely agnostic from an economic viewpoint in
(1) The revised Code of Conduct (Specifications on the
implementation of the Stability and Growth Pact and Guidelines on the format and content of Stability and Convergence Programmes) was endorsed by the Council in September 2005.
(2) Commission services' economic assessment of the
2006/07 vintage include short paragraphs comparing output gap estimates with the indications emerging from complementary indicators. The assessments can be found at:
http://ec.europa.eu/economy_finance/netstartsearch/pdfs earch/pdf.cfm?mode=_m2
the sense that it does not require any a priori as regards the link between the indicators selected and the cyclical conditions (3). It is organised in
two steps: The principal components are first calculated using the battery of complementary indicators and selected using statistical criteria; these components are then used to forecast the output gap using an ordinary least square regression.
The second approach to take into account complementary indicators is an extension of the current version of the commonly agreed production function method for calculating potential output and the output gap currently used in the EU fiscal surveillance framework, in order to incorporate the degree of utilization capacity of labour and capital.
One of the major difficulties in the commonly agreed method is to correctly identify total factor productivity (TFP), which generally represents the largest part of GDP growth. Currently, this is achieved by resorting to the simplifying assumption that the existing stocks of capital and labour are always fully utilised across different phases of the cycle. The price paid for the simplification is straightforward. To the extent that the degree of capacity utilisation increases during upswings and decreases in downswings, TFP may be over- or underestimated, which in turn may affect the accuracy of the output gap estimate in real time.
One way to overcome the problem is to relax the simplifying assumption about the constant degree of utilisation of capital and labour and to make use of available survey data on the rate of capacity utilisation and to embed it into the commonly agreed model, so as to track the
(3) The technical details of how forecasts are made based
on Stock and Watson (2002) approach are explained in Annex II. The properties of this forecasting procedure are based on the fact that the information potentially available in a large number of variables over long time spans provides valuable statistical information that can be exploited for forecasting purposes without imposing any a priori restriction on the links among these variables and the variable to be forecast.
Box II.2.1: How the budget balance is adjusted for cyclical factors in the EU fiscal surveillance framework
In the framework of the EU budgetary surveillance the cyclically-adjusted budget balance (CAB) is derived by subtracting the temporary component of the budget balance from the overall nominal figure:
t OG t b t CAB= −ε⋅ (1)
where bt is the nominal budget balance-to-GDP ratio in year t,
ε
the budgetary sensitivity parameterand OGt the output gap in year t. The output gap is derived from a production function method endorsed
by the Council in July 2002. A detailed description of the method can be found in Denis et al. (2006). The
overall sensitivity parameter
ε
is obtained by aggregating the elasticities of individual budgetary itemsestimated on the basis of the methodology developed by the OECD and agreed by the Output Gap
Working Group of the Economic Policy Committee (OGWG) (1). The individual revenue elasticities,
i R,
η , are aggregated to an overall revenue elasticity ηR using the share of each in the total current taxes
(Ri/R) as weight: ( 2) . 4 1 , ∑ = = i i i R R R R η η (2)
As for the expenditure elasticity, ηG, it can be expressed as
G GU U G G η, η = (3)
where ηG,U is the elasticity of unemployment-related expenditures, again estimated on the basis of the
agreed OECD methodology, and GU /G is the share of unemployment related expenditure in total
current primary expenditure (3).
The empirical estimates of the individual tax and expenditure elasticities for all EU Member States are reported in Table 1 together with the overall tax and expenditure elasticities. The weights used to aggregate the elasticities of the individual tax categories are shown in Table 3.
As budgetary variables are generally expressed in percent of GDP, the revenue and expenditure
elasticities ηR and ηG (which measure the change in the level of a budgetary item with respect to the
output gap) are transformed into sensitivity parameters as follows:
Y G Y R G G R R η ε η ε = , = , (4)
(1) The OECD method for estimating budgetary elasticities is described in detail in Girouard, N. and C. André (2005).
(2) The weights are computed by the Commission services as an average over recent years. The period over which the
average is computed for the new and updated values of the budgetary elasticities is 1995-2004 (or 1995-2003 in case 2004 was not available)
(3) The share is computed by the Commission services using OECD data or data from national source for non-OECD
countries. The reference year is 2003 (2002 if not available).
variations in the use of the existing capital stock during up- and downswings periods (1).
Both methods − the principal components approach and the extended production function −
have their pros and cons. The clear advantage of the extended production function method is that it stays within the commonly agreed method for calculating potential output and the output gap in the EU fiscal framework. The inclusion of the rate of capacity utilisation does not alter the overall philosophy of the agreed approach, which, due to its formal status, plays a pivotal role in the implementation of the SGP (2). The
downside of the extended production function approach is equally clear. On top of the rate of capacity utilisation there may be other data available in real time (e.g. current account balance, asset price developments) that encapsulate useful information about the economy's position in the cycle. Such additional variables do not fit into the commonly agreed method, however. By contrast, they can be included in the principal components approach. Indeed, the agnostic feature of the principal components analysis does not set any limits to the type and number of variable to be used. But the potentially all-embracing nature of this latter approach comes at a price: the results of the principal components are rather opaque in the sense that it is difficult to pull out an economic story of which variables actually account for the predicted position in the cycle. This method also
(1) The technical details of the augmented production
function approach are provided in Annex II.
(2) In July 2002 the Council endorsed the production
function approach as the reference method for the calculation of potential output and the output gap in the EU fiscal surveillance framework.
relies on the availability of a large number of complementary indicators over long time spans which, for instance in the case of the recently acceded Member States, poses a problem.
A numerical simulation
In order to evaluate the relative merits of the two approaches in practice, a simulation for twelve EU countries was carried out covering the years 2000-2006 (3). The objective of the simulation is
to find out whether the two methods described above improve the assessment of the cycle in real time compared to the commonly agreed method currently applied in the EU fiscal surveillance framework. To this end the simulation replicates a typical assessment situation in the EU surveillance framework; i.e. only data available at the moment of the Commission services' assessment of the stability and convergence programmes are used (4). The latter is ensured by
using successive vintages of the Commission services' autumn forecasts available at a given year t. The benchmark for the 'true' output gap is the one estimated with the production function approach on the basis of the latest available information (i.e. at the time the present report was written), the Commission services' 2008 spring forecasts − henceforth ex-post output gap. The accuracy is assessed by means of graphs comparing the alternative real-time estimates and a basic statistical analysis. In view of the limited
(3) The twelve countries are BE, DE, DK, GR, ES, FI, FR, IE,
IT, NL, PT, UK were selected on the basis of the data availability of the rate of capacity utilisation variable. The first year available is 2000 since this year also corresponds to the oldest vintage of real-time estimates based on the commonly agreed production function approach.
(4) The simulation design is detailed in Annex II. Box (continued)
where R/Y is the share of current taxes in GDP and G/Y is the share of primary current expenditure on
GDP. (1) The difference
G R
ε
ε
−
yields the sensitivity parameter of the overall budget balanceε
used in equation (1).
sample length the statistical quality of the accuracy analysis is likely to be relatively poor. Nevertheless, the comparison of alternative estimates provides useful insights, especially for years when the commonly agreed method proved to be wide off the mark, for instance in the year 2000 and 2001. Table II.2.1 provides two measures of the accuracy of the real-time output gap forecasts, namely the Mean Error (ME) and Mean Absolute Error (MAE) for the three alternative methods: the commonly agreed production function approach, the commonly agreed production function approach incorporating the capacity utilisation and the method based on the principal components approach (1).
Table II.2.1 shows that using methods other than the commonly agreed production function provides additional useful information, although it must be said that no method clearly outperforms. On average, the extended production function approach including the capacity utilisation gets the best scores, while the principal components tends to do a better job in certain cases although this depends on the statistical criteria used. Generally speaking, the use of complementary indicators tends to
(1) The ME measures the mean of the difference between
the output gap estimates using the three different approaches and the ex-post estimates taken from the Commission services' spring 2008 forecasts for the period 2000-2006. The MAE takes the mean of the absolute values of the differences instead.
improve the forecast of the output gap for a number of countries. When using the MAE criteria, the incorporation of the capacity utilisation improves the real-time estimates for Germany, Denmark, France, Italy, the Netherlands and Finland. For the latter two countries, however, the current production function is equally performant. In addition, the current production function approach yields the best results for Portugal, Ireland and the UK. The use of the principal components approach improves the real-time estimates for Belgium, Greece and Spain. The same proportions roughly applies when using the ME criteria instead, although the countries' grouping changes slightly.
The values of the ME statistic show that in a majority of cases the production function approach incorporating the rate of capacity utilisation yields real-time output gap estimates which are larger than the real-time estimates based both on the commonly agreed method and the principal components approach over the period considered (2000-2006). For the EU fiscal surveillance, this would imply an additional margin of caution in order to take into account the uncertainty attached to the assessment of the cyclical component of the cyclically-adjusted balance.
The relative merits of each method may vary depending on the country considered. As mentioned above, the real-time output gap estimates differed substantially from the ex-post
Table II.2.1
Accuracy statistics of real time output gap estimates for 2000-2006. Benchmark: Commission services' spring 2008 forecast
MAE ME MAE ME MAE ME
Belgium 0.67 -0.67 0.85 0.85 0.55 -0.47 Germany 0.87 -0.47 0.69 0.21 1.01 -0.53 Denmark 0.86 -0.35 0.81 -0.73 0.83 -0.37 Greece 0.93 0.93 1.42 0.94 0.79 0.74 Spain 1.17 -1.00 1.49 -1.49 1.03 -0.86 Finland 0.46 0.19 0.46 0.19 0.88 0.59 France 0.86 -0.86 0.23 0.08 0.70 -0.70 Ireland 1.48 -0.95 4.17 -4.17 1.48 -0.74 Italy 1.43 -1.43 0.76 0.57 1.27 -1.27 Netherlands 0.71 -0.66 0.71 -0.15 0.79 -0.76 Portugal 1.52 -1.52 1.63 -1.60 1.71 -1.71 UK 0.76 -0.76 1.62 -1.62 0.84 -0.84
Source: Commission services.
Commonly agreed production function approach
Commonly agreed production function approach incorporating the capacity
utilisation
Complementary indicators, principal component
Notes: ME=Mean error; MAE = Mean absolute error. The benchmark output gap series is the one based on the Commission services' spring 2008 forecast. Figures in bold correspond to the best performing forecast.
values observed in the early 2000s. A comparison of the three methods for countries where forecast errors during these years turned out to be particularly large can be particularly useful. Three countries, namely the Netherlands, Italy and Greece were selected on the basis of the average forecast errors for the years 2000 and 2001. Results are reported in Graph II.2.1 to Graph II.2.3.
Graph II.2.1:Assessment of cyclical conditions using alternative methods: Netherlands, 2000-2006
-3 -2 -1 0 1 2 3 4 2000 2001 2002 2003 2004 2005 2006 Output gap, expost
Principal components, realtime Poduction function approach, realtime Extended production function approach, realtime
Source: Commission services
Graph II.2.1 shows that, in the case of the Netherlands, the extension of the production function approach to account for the capacity utilisation overall provides better predictions compared to the two other methods. This result is