European commission’s fiscal forecasts in CEE countries: a thorough assessment

ABSTRACT

This paper assesses European Commission’s fiscal forecasts for a sample of 10 Central and Eastern European countries between 2005 and 2015. The analysis focus on forecasts of the budget balance, revenues, expenditures and debt and pays special attention to dynamics around business cycle turning points. Results suggest that the distribution of projection errors appears to be biased towards optimism of fiscal aggregates and accuracy increases as the forecast horizon shortens. We also find evidence of “forecast smoothing”. In addition, we find that, on average, the extent of optimism seems to increase during recessions (and to a lesser extent during recoveries). Moreover, errors in forecasting fiscal variables can be explained by forecasts errors of real GDP growth and inflation.

KEYWORDS:

• European commission forecasts
• Central and Eastern Europe
• turning points
• rational forecasts
• efficient forecasts

JEL CLASSIFICATIONS:

• C53
• E27
• E37
• E62
• D8

Acknowledgements

We thank the editor and two anonymous referees for useful comments and suggestions. The usual disclaimer applies. All remaining errors are the author’s sole responsibility.

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

1. Since national governments express their views about the outlook for fiscal policy in the form of annual targets and plans rather than projections or forecasts, the activities of revenue estimation and spending planning are key in the elaboration of annual budgets and the determination of (multi-annual) targets.

2. Sometimes this fact has led to claims that public finance projections in Europe should be produced by independent authorities (Fatas et al. 2003; Merola and Perez, 2013).

3. Jonung and Larch (2006) claim that in some Euro area countries biased forecasts have played an important role in generating excessive deficits in the past.

4. Nevertheless, Brück and Stephan (2006) challenged EC projections for presenting several shortcomings, including the correlation of forecast errors with political cycles.

5. An overwhelming majority of the existing papers dealing with short – and medium-term fiscal projections or attempting to assess the effect of the cycle on the budget focus on government revenues, in particular on important items such as Wage Taxes, Sales Taxes or Value Added Taxes, and Social Contributions (Lawrence, Anandarajan, and Kleinman 1998; and Van den Noord 2000). However, it is not unusual to see short-term forecasting models of the spending side of the budget as, for example, in Mandy (1989) for unemployment insurance funds, or Tridimas (1992), Giles and Hall (1998), Sentance, Hall, and O’Sullivan (1998) and Pike and Savage (1998) for an integrated view of both revenue and spending sides.

6. We are aware that, despite the merits of EC forecasts (Keereman 1999), the use of private sector forecasts would be preferable as they are known to be hard to beat (Batchelor and Dua 1992) and potentially less contaminated by political economy considerations. However, due to the absence of private sector fiscal forecasts covering CEE countries (such as Consensus Economics data), we have to rely on EC’s forecasts. We thank an anonymous referee for this point.

7. Even though the European Commission has started publishing forecasts in 1998, they only began to do so for CEE countries much later, in 2005. This fact dictates the beginning of our time span.

8. We use the autumn issue only since the spring one for CEE countries is not complete throughout the entire time span we cover.

9. At this point a caveat should be made. Data revisions have worried economists for many years now (see Croushore 2011 for a recent survey) and policy-makers have to base their decisions on preliminary and partially revised data, since the most recent data are usually the least reliable as it simply translates a noisy indicator for final values (Koenig et al., 2003). We are aware that data revisions pose challenges to forecasters and that forecasting studies should reflect the true forecasting performance by using real-time data instead of final data (Stark and Croushore 2002). However, in the present case it is difficult to find reliable, consistent and comparable real-time vintages for the set of fiscal variables used in the paper and the sample considered.

10. Note that the main point of Burns and Mitchell (1946) was that business cycles are fluctuations of aggregate economic activity and not just that of a single variable. Due to the difficulty in accounting for a multivariate approach empirically in our present setting, we employ this simple approach which similar to NBER’s definition for dating turning points in the US business cycle. See Claessens, Kose, and Terrones (2008) for a discussion on business cycle dating in advanced economies. We thank an anonymous referee for this point.

11. Results obtained by plotting instead a Kernel density at each horizon give a qualitatively similar overall picture (not shown). We used the Epanechnikov kernel, an inverted-U quadratic curve, which is actually the most efficient kernel and a common choice in econometrics.

12. Defined as the average difference between the actual value and its forecasted value. For example, a positive value for bias in the case of the budget balance, indicates that on average over the whole run of forecasts, the actual value was under-estimated, or that the forecasts were too low.

13. The RMSE may be the most popular measure among statisticians partially because of its mathematical tractability. More recently, researchers seem to prefer the so-called Percent Better, the Mean Absolute Percentage Error and the Relative Absolute Error. For a review see Armstrong and Collopy (1992), Fildes (1992) and Baillie, Bollerslev, and Mikkelsen (1993).

14. Berger, Kopits, and Szekely (2007) found evidence of a significant loosening in the Czech Republic and Hungary after 1999 coincident with these countries accession to NATO.

15. That said, inspecting and studying such institutionally driven determinants of fiscal forecast errors go beyond the scope of this paper. We do try to justify fiscal forecast errors with GDP and inflation forecast errors in section 4.

16. As a rule, if forecasts are in line with the Rational Expectations Hypothesis (REH) formulated by Muth (1961), they should be unbiased. The REH states that market participants use all cost-efficient knowledge to forecast.

17. Note than since forecast errors might be prone to serial correlation, in order to deal with this, we have employed Newey-West’s correction and present heteroskedasticity and autocorrelation robust standard errors.

18. It is worth pointing to the possibility that some of the rejections of forecast optimality may simply be driven by the assumption of MSE loss rather than the absence of forecast rationality per se. Elliott et al. (2005) state that MSE loss, despite a generally used assumption, is often hard to justify on economic grounds and is subject to debate and criticism. In fact, asymmetric loss captures the idea that the cost of over – and underpredicting a given variable may be very different. Patton and Timmermann (2007) propose a transformation of the forecast error that possesses the same set of rationality properties under asymmetric loss and nonlinear data generating processes.

19. The literature has identified three main potential explanations for “forecast smoothing” or “information rigidity” (for departure from FIRE): i) behavioral explanations (Nordhaus 1987; Tversky and Kahneman 1981); ii) sticky-information (Mankiw and Reis 2002) and iii) imperfect information (Woodford 2002; Sims 2003).

20. Mankiw and Reis (2002) propose a model of inattentive agents who update their information sets each period with probability (1-λ), but acquire no new information with probability λ, so that λ can be interpreted as the degree of information rigidity and 1/(1 – λ) is the average duration between information updates. In the context of sticky information models, λ = β/(1 + β), with β being the estimated rigidity coefficient.

21. One feature of this test is that it requires the use of the actual realizations and hence requires a view on whether to use the latest data or an earlier vintage. Our tests use the latest data.

22. We thank an anonymous referee for making this point.

23. http://www.imf.org/external/np/fad/council/.

24. See Gaspar, Obstfeld, and Sahay (2016) for a recent discussion on the topic and on the “3Cs approach” to economic policy.