A PANIC analysis on regional and sectoral inflation: the Spanish case

Abstract

This article studies the stochastic properties of several inflation rates for the Spanish economy using the consumer price index (CPI) for the 17 regions and 12 groups of goods and services, and the producer price index (PPI) for 26 industrial sectors. To this end, we employ the panel analysis of nonstationarity in idiosyncratic and common components (PANIC) approach proposed by Bai and Ng (2004, 2010). This methodology enables us to decompose the observed inflation rate series into a common and an idiosyncratic component, thus allowing us to identify the exact source of nonstationarity. Our analysis provides strong evidence of the presence of a common stochastic trend driving the observed series forming the panel of CPI-based inflation rates for the regions. This, coupled with the presence of a jointly stationary idiosyncratic component, implies the existence of pairwise cointegration across the regional CPI-based inflation rates, which show a clear pattern of convergence over time. This gives an indication of increased geographical homogeneity in consumption patterns. The evidence for the panels of CPI-based inflation of groups of goods and services and PPI-based inflation of industrial sectors indicates the existence of four independent common stochastic trends. This, combined with jointly stationary idiosyncratic series, provides much weaker evidence of cross-cointegration for these two panels.

Keywords:

• inflation
• PANIC methodology
• consumer and producer prices
• sectoral and regional analysis
• common stochastic trends

JEL Classification:

• C23
• E31
• E52
• E58
• R10

Acknowledgements

We thank Mª Ángeles Caraballo and Daniel Oto-Peralías, as well as participants at the 14th INFER Annual Conference (Pescara, 2014) and the XL Reunión de Estudios Regionales (Zaragoza, 2014), for valuable comments and suggestions. The authors are particularly indebted to the Editor and two anonymous referees of this journal for valuable suggestions that led to a substantial improvement of the original manuscript. The usual disclaimer applies.

Notes

¹ In the extreme case in which there are no cointegrating vectors, there would be N independent common stochastic trends, and no evidence of cross-cointegration and convergence. At the other end, if there are no common stochastic trends (${\hat{r}}_1=0)$, it implies that there are N cointegrating vectors, thus indicating that all common factors are I(0) and the individual series are stationary (Gengenbach et al., 2009, p. 128).

² Note that the statistical justification for using PANIC goes hand in hand with the economic intuition of the technique and the policy implications that one can derive from the results. The key is the decomposition of the observed series into a common and an idiosyncratic component, and in turn the determination of their degree of integration, which allows us to establish whether idiosyncratic and/or common shocks or policies can have transitory or permanent effects on inflation, depending on the degree of integration of either component.

³ Existing low labour mobility can also contribute to slowing down the transfer of resources between sectors (Jimeno and Bentolila, 1998).

⁴ The reasons that account for this phenomenon are an overall reduction in competition in many markets during the Great Recession and the existence of restrictive financial conditions that exert upward pressure over the markups. A composition effect stemming from the exit of the less productive firms, whose markups are lower than the more productive ones’, might have also contributed to bidding up average markups.

⁵ For instance, the Inflation Persistence Network (IPN) was a team of researchers from the ECB and other Euro-system national central banks, who, over the 2000s, conducted a coordinated project on the patterns, determinants and implications of inflation persistence in the Euro-area and in its member countries. This team was tasked with producing reports on the aforementioned subject. Although it is no longer active, it is worth pointing out that within the Euro-system, other existing research networks on related issues are now playing a similar role, such as the Wage Dynamics Network (WDN). Individual papers employing unit root tests other than PANIC to test the hypothesis of a unit root in inflation for different groups of countries include, among others, MacDonald and Murphy (1989), Charemza et al. (2005), Basher and Westerlund (2008), Narayan and Narayan (2010), Caporale and Paxton (2013) and Zhou (2013).

⁶ Two additional measures of inflation persistence also linked to the ‘hybrid’ NKPC are the indicators related to the state of the cycle or ‘explicit persistence’ – a concept that includes determinants of current inflation such as labour costs or the firms’ real marginal costs and the output gap, and also captures rigidities in the wage- and price-setting processes – and the inflation expectation component – which comprises the backward- and forward-looking approaches.

⁷ Galí and López-Salido (2001) and Fabiani et al. (2006) stress the importance of the backward-looking component in Spain in accounting for inflation persistence relative to the Euro-zone average. The former also finds wage frictions to be a relevant explanation of this phenomenon. On the other hand, Restoy et al. (2005) highlight the prominent role of wage indexation clauses as well as the dual inflation problems. Caraballo and Usabiaga (2009b, c) and Caraballo and Dabús (2013) present robust evidence for the presence of nominal rigidities in the determination of Spanish consumer and producer prices. Finally, Caraballo and Usabiaga (2009c) and Álvarez et al. (2011) also emphasize the vulnerability of Spanish inflation to oil shocks.

⁸ Álvarez and Hernando (2006), Álvarez et al. (2010) and Romero-Ávila and Usabiaga (2012), focusing on the Spanish case, arrive at this result.

⁹ It is customary to confront European (and Spanish) results with USA’s, a country often taken as the benchmark as far as price flexibility is concerned.

¹⁰ This representation corresponds to the factor model with a constant. For the representation in the case of a specification with a trend, we refer to Bai and Ng (2004b, p. 1137).

¹¹ The second argument in the loss function stands for the penalty for overfitting, which is intended to correct for the fact that models with a larger number of factors can at least fit as good as models with fewer common factors, but efficiency is reduced with the estimation of more factor-loading parameters (Bai and Ng, 2002).

¹² The use of unit root statistics (for the case of testing the unit root null hypothesis) in tandem with stationarity statistics (for the case of testing the stationarity null hypothesis) enables us to conduct a confirmatory analysis on the stochastic properties of the inflation rate series. Therefore, when the null hypothesis is rejected with the panel stationarity test but not with the panel unit root test, all cross-sectional units contain a unit root; and when there is rejection with the panel unit root test but not with the panel stationarity test, all cross-sectional units are I(0). In addition, when the respective null hypotheses in both panel unit root and stationarity tests are rejected, this indicates the existence of a mixture of stationarity and nonstationarity in the panel, whereas if the respective null hypotheses in both tests are not rejected, the results would be inconclusive in all likelihood due to the poor information provided by the data set. See more details in Shin and Snell (2006, p. 136).

¹³ The asymptotic distribution of $AD{F}_{\hat{e}}^c(i)$ is the same as the Dickey–Fuller distribution for the case of no constant, while that of $AD{F}_{\hat{e}}^{\tau }(i)$ is proportional to the reciprocal of a Brownian bridge. A Brownian bridge is considered a continuous-time stochastic process B(t) with a probability distribution equalling the conditional probability distribution of a Wiener process W(t) (which can in turn be considered a mathematical model of Brownian motion), given the condition that B(1) = 0. More specifically, $B_t=(W_t\left(W_1\right)=0){,}_{\hspace{0.17em}}t\in \left[0,1\right]$, with the expected value of the bridge being equal to zero, with variance equalling t(1 − t), which implies that uncertainty is the highest in the middle of the bridge, while zero at the nodes.

¹⁴ The same holds for the case of a trend, where ${\pi }_{\hat{e}}^{\tau }(i)$ is the p-value associated with $AD{F}_{\hat{e}}^{\tau }(i)$. The pooled statistics for the trend specification are denoted as $P_{\hat{e}}^{\tau }$ and $Z_{\hat{e}}^{\tau }$. Note that we do not pool individual unit root tests for the observed series, since under a factor structure, the limiting distribution of the test would contain terms that are common across units. However, ‘pooling of tests for ${\hat{e}}_{it}$ is asymptotically valid under the more plausible assumption that ${\hat{e}}_{it}$ is independent across i’(Bai and Ng, 2004b, p. 1140).

¹⁵ The determination of the number of stochastic trends in the system follows a sequential testing procedure, in which we first assume that the rank of the cointegrating space is zero, that is, the number of stochastic trends is equal to that of common factors (m = k). We then specify the null hypothesis that there are m stochastic trends against the alternative hypothesis of less than m common stochastic trends. If the null hypothesis is rejected, we then specify the null hypothesis of m − 1 stochastic trends and test it against the alternative of less than m − 1 common stochastic trends, and we continue this process until we fail to reject the null hypothesis or when m = 0 is achieved, in which case, there are no common stochastic trends. The critical values of the $M{Q}_f$ and $M{Q}_c$ rank statistics are provided in Table I in Bai and Ng (2004b). See Carrión-i-Silvestre and Surdeanu (2011) for a similar procedure, but under panel cointegration.

¹⁶ The PPI measures the prices of industrial products, sold in the domestic market, during the first step of its commercialization, excluding transport costs, merchandising and indirect taxes. The index coverage extends to all industrial sectors, except construction. It therefore comprises extractive industries, manufacturing and supply of electricity, gas and water.

¹⁷ Romero-Ávila and Usabiaga (2009, 2012) tackle a similar topic, but the former uses the aggregate inflation figures of a set of OECD countries and employs panel data techniques – for comparison purposes, they also apply univariate tests of the KPSS type, obtaining evidence of a unit root for Spain, which coincides with the results in this article and of Romero-Ávila and Usabiaga (2012) for aggregate CPI inflation. The latter also makes use of several univariate tests, reaching the same conclusion of high persistence for different Spanish inflation series.

¹⁸ From a more microeconomic viewpoint, the IPN has generated publications for the Euro-zone and member countries, but employing empirical methodologies different from ours.

¹⁹ As will become apparent below, the main results are fairly robust to the inclusion of a linear trend in the specification.

²⁰ In order to provide additional evidence in this respect, it should be noted that Caraballo and Usabiaga (2009a) analysed the weights set in 1992 and 2009 concerning the three groups with the highest CPI weights in the consumption basket in 2009 – G1 (food and nonalcoholic beverages), G7 (transport) and G11 (restaurants, cafés and hotels), in this order. They derived the SD of the weights for these groups in each region in 1992 and 2009 relative to the nation-wide weight. The SD decreases over time for these groups, which points to greater homogeneity in the weights. This result can be thought of as another indication of increased geographical homogeneity in the consumption patterns exhibited by consumers in the different Spanish regions. In contrast to regional CPI inflation, we find no clear-cut regional convergence patterns as regards the PPI series (with such series being only available since 2003, which prevented us from applying the PANIC methodology to these series). This indicates that there is more heterogeneity in the patterns of production (than in those of consumption) across regions, as reflected in the fact that regions are not specializing in the same manufacturing and energy products, whose prices are covered in the PPI series.

²¹ Caraballo and Dabús (2013) argue that another problem facing the Spanish economy is that the ECB inflation target does not coincide with the average inflation minimizing the RPV. This would be in contrast to the case of other neighbouring countries, which would not suffer from this predicament (Caraballo and Efthimiadis, 2012).

²² The only exception to this is given by the failure to reject the joint nonstationarity null hypothesis for the panel of CPI-based inflation rates of groups of goods and services with the pooled panel unit root tests of Bai and Ng (2010). This should not come as a surprise, since the removal of the years covering the Great Recession leaves a relatively short time span in this case.

²³ The small number of years forming the period of the Great Recession prevented us from applying PANIC to it.

²⁴ Despite the fact the crisis has hit Spain hard since 2008, some urgent measures have been delayed a great deal. For example, it was not until October 2013 that the final organization of the Comisión Nacional de los Mercados y de la Competencia, responsible above all for the application of the Act 15/2007 of Defense of Competition, took place. By the same token, it was not until December 2013 that the Act of Garantía de la Unidad de Mercado was enacted, with the aim of lowering geographical barriers to competition. On both fronts, only tentative progress has been made thus far.

Additional information

Funding

This study was supported by the Spanish Ministry of Science and Technology [grant number ECO2009-13357]; the Spanish Ministry of Economics and Competitiveness [grant number ECO2012-35430]; and the Andalusian Council of Innovation and Science under Excellence Project SEJ-4546.