The cyclicality of consumption, wages and employment of the public sector in the euro area

Abstract

This study empirically examines the business cycle behaviour of public consumption and its main components, the public wage bill (including its breakdown into compensation per employee and public employment) and intermediate consumption, in the euro area aggregate, euro area countries and a group of selected non-euro area Organization for Economic Co-operation and Development (OECD) countries (Denmark, Sweden, the UK, Japan and the US). It looks across a large number of variables and methods, using annual data from 1960 to 2005. It finds robust evidence supporting that public consumption, wages and employment co-move with the business cycle in a pro-cyclical manner with 1–2 year lags, notably for the euro area aggregate and euro area countries. The findings reflect mainly the correlation between cyclical developments, but also point to an important role of pro-cyclical discretionary fiscal policies.

Keywords:

• public wages
• public employment
• detrending
• business cycle
• stylized facts

JEL Classification::

• E62
• E63
• H50

Acknowledgements

The views expressed in this article are the authors' and do not necessarily reflect those of the European Central Bank or the Bank of Spain. We thank two anonymous referees, Fabio Canova, Henri Maurer, Matthias Mohr, Alfonso Novales, Evi Pappa, Diego Pedregal, Ad van Riet, Jürgen von Hagen, and participants in the Workshop on International Economics (Valencia, Spain), Ecomod (Sao Paulo, Brazil), EEA (Budapest, Hungary), University Pablo de Olavide (Seville, Spain) and WDN (Frankfurt am Main, Germany) for their comments.

Notes

¹ The theoretical literature suggests little consensus as to whether fiscal policies should have, are likely to have, or in fact do have a stabilizing effect on demand. Keynesian economics suggests that governments should and would stabilize demand by behaving counter-cyclically while the normative predictions from a neoclassical perspective depend on the relationship between private and public consumption. Political economy models generally predict pro-cyclical discretionary policies, particularly in developing economies, e.g. Akitoby et al. (2009), Lane and Tornell (1999), Schuknecht (1999) or Talvi and Vegh (2005). Gavin and Perotti (1997), Riascos and Vegh (2003) and Mendoza and Oviedo (2006) among others, also predict pro-cyclicality steaming in this case from market imperfections. Some grounds for a-cyclicality can be found in political economy models in which the different status that civil servants enjoy might make public wages and consumption less reactive to the business cycle, or even generate a separate agenda of public employees (rent-seeking behaviour), as in Borjas (1984) or Matschke (2003). Klomp and de Haan (2009) find that democracy often reduces fiscal policy uncertainty which, in turn, reduces economic volatility. Interest groups see public spending (Lane and Tornell, 1996) or taxation (Alesina et al., 2008) as a common good to deviate to their benefit, and the pressure to do so, is stronger in economic booms. It seems plausible that public employment is an important way to distribute rents.

² Euro area countries in this study include Belgium, Germany, Greece, Spain, France, Italy, the Netherlands, Ireland, Austria, Portugal and Finland.

³For data sources and variable definitions refer to Appendix I.

⁴ Even though the size of government consumption over GDP is similar in Germany and other euro area countries, the composition of government consumption is somewhat different. Compensation of employees makes up around 40% of total government consumption in Germany against 55% in France, Italy or Spain. At the same time, the item ‘social transfers in kind provided via nongovernment units’ (reimbursements of social security benefits and social assistance benefits) amounts to 40% of government consumption in Germany, compared to 25% in France and Italy or 14% in Spain. The case of the Netherlands is similar to that of Germany.

⁵ The interpretation of intermediate consumption might be somewhat biased in the case of the UK. UK National Accounts do not contemplate a formal social security sector, and thus a great deal of expenditure items under ‘social transfers in kind provided via nongovernment units’ are recorded under intermediate consumption. This explains that this latter item amounts to some 50% of total government consumption in the UK statistics.

⁶ Note that the relative public–private sector wages ratios can be affected by a different occupational composition in the public versus private sector.

⁷ See, for example, Abraham and Haltiwanger (1995) that offers a review of the debate on wage cyclicality. They report pro- or counter-cyclicality depending on the measure of real wages used and on the business cycle indicator.

⁸ The annual data set used is the same (but more up-to-date) than the one used by Lane (2003) which is a close reference to ours. In addition the availability of quarterly data for government wages, government employment and intermediate consumption is extremely limited. For most European countries the data is not publicly available and for those for which some nominal, nonseasonally adjusted time series are available, the typical sample only runs from 1990 to 1995 onwards, and not for all variables. For some details on the availability of quarterly fiscal data for EU countries the interested reader is referred to European Commission (2007 pp. 103–6).

⁹ See, among others, Agenor et al. (1999), Stein et al. (1999), or Talvi and Vegh (2005).

¹⁰ Empirical evidence in Blanchard and Perotti (2002) and other article in the Structural Vecto Autoregressive (SVAR) literature that have followed upon that seminal article, shows that across countries and time periods, rises in public spending are likely having expansionary effects on output.

¹¹ See for example André et al. (2002), André and Pérez (2005), Box et al. (1994) or Bartlett (1955).

¹² TRAMO stands for ‘Time Series Regression with ARIMA noise, Missing Values and Outliers’. See Gómez and Maravall (1996).

¹³ In the somewhat related literature on meta-analysis the basic principle is to calculate effect sizes for individual studies, convert them to a common metric, and then combine them to obtain an average effect size (see Rosenthal, 1991). In an application to economics Camacho et al. (2006) also use Fisher transformations to combine correlation coefficients.

¹⁴ In our case T _i  = T, ∀t.

¹⁵ See for example Kydland and Prescott (1990) and Fiorito and Kollintzas (1994).

¹⁶ The cutoff point for the combined correlation in the case of combining independent correlation coefficient estimates, which is not strictly our case, would be slightly above 0.1. Nevertheless, some studies recommend (see Rosenthal, 1991) to calculate the probabilities for combined correlation by combining the individual probability values of each correlation coefficient, in which case our cutoff point would be close to 0.3. We take 0.2 as a compromise between the two alternatives which is in line with the cutoff values normally used in the literature.

¹⁷ There is however an identification issue in interpreting the correlation as strictly due to discretionary measures. For example, if an economic boom is accompanied by higher employment (with lags) in the private sector, this may raise private real wages. If there are linkages between the private and public sector, such wage rises may spill over (with a lag) to the public wage bill. This may induce procyclical behaviour in fiscal policy, without the need for discretionary fiscal measures.

¹⁸ We computed the standard Jarque–Bera normality test for the distribution of each set of z ₁,…, z _N . This amounted for each country, to one sequence per considered correlation in the CCF (−2, −1, 0, 1, 2) and per pair of variables considered. In most cases (on average more than 90% of the cases) the null hypothesis of normality was not rejected.

¹⁹ Similar tables for each country are available from the authors upon request.

²⁰ All models were estimated by exact maximum likelihood using the MATLAB toolbox of Pedregal (2004).

²¹ Pre-whitening has been traditionally performed with ARIMA specifications, but the basic concept applies to any other econometric representation. An ad hoc pre-whitening procedure could be designed to address any econometric setting, depending on the available information.