Euro area inflation: long-run determinants and short-run dynamics

Abstract

This study adopts the long-run structural VAR approach to analyse the determinants of inflation in the Euro Area economy over the period 1985:1–2003:2. Theoretical relationships link inflation to markup and output gap, respectively. The short-run dynamic properties of inflation are investigated using a structural VECM. Inflation is explained by a mixture of supply- and demand-side factors, both in the long- and the short-run. The simulation exercise indicates that a positive shock to inflation could have a favourable re-distributional income effect on wage earners and non-detrimental consequences either on productivity and on competitiveness. Finally, the model produces satisfactory out-of-sample forecasts.

Acknowledgements

The authors are grateful to Alberto Bagnai, Francesco Carlucci, Marcus Chambers, Giuseppe De Arcangelis, Aditya Goenka, Paolo Paesani and Adrian Pagan for helpful comments and suggestions on a previous version of the paper. A special thank to Lucio Sarno for comments and discussions. The views expressed do not necessarily reflect those of the Ministry of Economy and Finance and of the Institute for Studies and Economic Analyses of Italy.

Notes

¹ Cabos and Siegfried (2004) investigate the performance of monetary policy strategies alternative to the ‘two pillars’ one adopted by the ECB.

² For an alternative approach to the Euro Area inflation see, for example, Bagliano et al. (2002).

³ Treating inflation as an I(1) process makes it hard to target by monetary policy. Using other data sets, like the ECB one, this does not seem a good characterization of the Euro Area inflation. We thank Adrian Pagan and Lucio Sarno for pointing this out. Nevertheless, the existence of a unit root in inflation is still a controversial issue (see, among others, Charemza et al., 2005).

⁴ The presence of Δp_t in Equation 1 implies that inflation may represent a cost to firms even in the long-run (e.g. because of the difficulties faced by price-setting firms in adjusting prices in an inflationary environment with incomplete information). Thus, an increase in costs may not be fully reflected in higher prices because the markup falls with higher inflation.

⁵ Alternatively, an algorithm for the extraction of trend from actual output or an explicit statistical model can be used (Harvey and Jaeger, 1993; Clark et al., 1996).

⁶ Under the assumption that the share of employed workers on population is stationary, as in Garratt et al. (2003), (labour) productivity may represent a measure of per-capita output.

⁷ This strategy can be justified from a statistical point of view since the presence in the model of variables erroneously treated as exogenous could produce non efficient estimates.

⁸ If r = 1, the above framework can also serve as a procedure to discriminate among competitive theories. If inflation is interpretable exclusively from a supply-side point of view, imposing an additional constraint to the r² = 1 exactly-identifying ones Equations 14 becomes:

From a demand-side perspective, Equations 14 becomes:

with two additional constraints.

⁹ Corresponding to the quarter 1995:3, as in the other recursive tests in this study.

¹⁰ The stochastic properties of the variables included in a VAR system can be investigated following two approaches. The first approach relies on the fact that the variables are part of a multi-equational model, which implies that the analysis should be conducted within a multivariate framework, as the one reported in the text. The second approach applies the unit root/stationarity test procedures to each series involved in the analysis. For the sake of completeness, Appendix presents the results of standard ADF tests as well as those of ‘augmented’ unit root tests to control for unknown breaks.

¹¹ Bowdler and Jansen (2004), using a different specification of the markup, estimate similar coefficients of unit labour cost and import prices.

¹² The estimated coefficient of inflation in Banerjee and Russell (2002b) is 4.925, but their specification of markup refers to a closed economy.

¹³ This level is close to that traditionally used by the Bundesbank (1.5%) to accommodate domestic structural changes (Sinn and Reutter, 2000).

¹⁴ A counter-cyclical markup may arise because of increasing marginal costs due to less productive inputs or to wages (more in general, to costs) pressures as economy expands (see, among others, Gali, 1994). Moreover, with markup and output gap negatively related in the long-run, the derivation of Equation 6 assumes that the linear homogeneity condition holds.

¹⁵ The choice is motivated by the opinion that, in the reduction process of the model, it is preferable to maintain the coefficients with uncertain significance.

¹⁶ This finding suggests the estimation of an analogous VECM in which the real exchange rate is treated as an I(1) exogenous variable. The trace and the maximum eigenvalue tests point out the existence of two cointegration relationships at the 5% and 10% significance level (with the critical values taken from Pesaran et al., 2000). The long-run structure in Table 1 is not rejected by the data (χ²(6) = 9.99 with a p-value of 0.125).

¹⁷ Forecast intervals are not drawn for better readability. However, the hypothesis that none of the forecasts are significantly different from the actual values is not rejected at the usual significance levels.

¹⁸ See Clements and Hendry (2001, p. 290).

¹⁹ Nevertheless, this result does not guarantee that the forecasts are significantly better than those of the benchmark model. The naïve model just establishes the minimum level of accuracy that system forecasts should have. Since the forecast exercise is based on a unique projection over the entire forecast horizon, standard tests on accuracy of rival forecasts (such as those proposed by Diebold and Mariano, 1995, or by Harvey et al., 1997) cannot be directly applied. When the Diebold and Mariano test is applied to static forecasts it indicates no substantial differences between ‘structural’ and naïve forecasts.

²⁰ The result of the independence test for the acceleration rate of price is not reported since at least one of the expected values is less than five. In this case, the test may be misleading.