Low and stable inflation during the period 1999-2008

8Average inflation in the euro area amounted to 2.2% on an annual basis during the period from January 1999 to September 2008. This is not only considerably lower than the inflation rates observed during the 1970s and 1980s, but also lower than the average inflation rate which was observed in the last 10 years before the start of EMU (period 1989-1998), that is, 3%. The euro area’s inflation performance since 1999 also compares favourably with the country that had experienced the highest degree of monetary stability during the four decades preceding EMU, namely Germany.

9However, euro area inflation was on average slightly higher than the 2% threshold in the quantitative definition of price stability. Such a positive deviation was particularly evident during the first nine months of 2008, when inflation was running at an average of 3.6%. This was mainly because of the sharp increase in prices for crude oil and other raw materials, like food commodities, in the second half of 2007 and the first half of 2008. As price increases of this type are highly unpredictable and have a very fast direct impact on headline inflation, they are very difficult to offset by monetary policy. Curbing these inflationary pressures, which are largely coming from abroad, could in fact lead to a sharp and undesirable increase in the volatility of both the policy instrument and domestic macroeconomic conditions. It is precisely to allow for a more gradual monetary policy response in such circumstances that an explicit medium-term orientation is embedded in the quantitative definition of price stability. This provides monetary policy with the possibility to accommodate the first-round effects of such changes in commodity prices, provided that the medium-term outlook for price stability is not clouded.

Figure 1

Inflation in the euro area and in Germany (percentage change compared to the previous year [1])

Inflation in the euro area and in Germany (percentage change compared to the previous year [1])

10Also, other temporary upward deviations of euro area inflation from the 2% threshold over the past ten years have been observed at times of strong oil price increases, when bad weather conditions or animal diseases sharply affected the food component of the HICP or when increases in indirect taxes or administered prices had a considerable impact on headline inflation.

11In this respect, it is striking to see that the HICP, excluding the more volatile food and energy components, not only displayed upward deviations from the 2% threshold on fewer occasions, but also generally remained below this threshold during the period 1999-2008. Note, however, that by focusing on this alternative inflation measure, one takes asymmetric account of the effects of globalisation. While this inflation measure excludes the direct upward influence of globalisation, via higher raw material prices, on the energy and food components of the HICP, it does not adjust for the downward pressure which it exerts on the prices of labour-intensive goods and services, particularly via imports of non-energy industrial goods from low-cost countries. Stressing that globalisation does exert two effects on inflation, which go in opposite directions, logically leads to the conclusion that this process essentially triggers significant changes in relative prices, rather than fundamentally affecting aggregate inflation. Obviously, these relative price changes can affect aggregate inflation in the short run and the sign of this effect may change over time, depending on which of both forces is dominant at a particular point in time. While monetary policy can accommodate these short-run effects, and in fact has done so, it fixes future inflation, essentially by steering demand in line with domestic supply conditions and by anchoring inflation expectations at a level consistent with price stability.

Inflation expectations anchored at levels consistent with price stability

12The Governing Council of the ECB has always stressed the forward-looking nature of monetary policy, in line with its medium-term orientation, and the importance attached to anchoring inflation expectations. Whenever these expectations started to diverge from levels consistent with price stability, for instance during the first half of 2008, particular concern was expressed, and, wherever deemed necessary, policy action was undertaken with the aim to curb inflation expectations. As a result, a remarkable degree of stability in inflation expectations, whether measured on the basis of surveys with economic forecasters or financial market instruments, was observed. This is particularly so when account is taken of the following two elements.

13First, the moderate upward drift in inflation forecasts 5 years ahead in the ECB’s Survey of Professional Forecasters (SPF) during the period 2002-2004 in fact results from a shift of the lower tail of the distribution, rather than from an upward shift of the entire distribution. The resulting concentration of forecasts between 1.9% and 2% could therefore be seen as being compatible with the below, but close to, two percent qualification of the definition of price stability, adopted by the Governing Council in May 2003.

14Second, when comparing financial-market-based break-even inflation rates (BEIR) with the 2% threshold in the definition of price stability, account should be taken of the fact that BEIRs are not only affected by genuine inflation expectations, but also by risk and liquidity premia, which are moreover time-varying. Under normal conditions, the upward impact of risk premia on BEIRs tends to dominate the downward impact of liquidity premia, which can explain why on average BEIRs have exceeded the 2% threshold, particularly at very long horizons. However, with the sharp intensification of the financial crisis that began in mid-2007, this does not need to be the case at the current juncture. As the related flight-to-quality considerations and preference for very liquid assets are much more pronounced for nominal bonds than for index-linked bonds, it could be that the liquidity premium has recently dominated the risk premium and therefore is putting additional downward pressure on BEIRs, over and above what the sharply deteriorated economic outlook would imply for genuine inflation expectations.

15By effectively stabilising inflation expectations, monetary policy has prevented a variety of forces, including globalisation, from exerting a lasting impact on inflation, implying that it is monetary policy that has ultimately pinned down the inflation rate in the medium to long run. This leads us to conclude that it is the creation of EMU, and in particular the implementation of the single monetary policy, rather than real forces like globalisation, which has brought about the high, and in recent decades unprecedented, degree of price stability in the euro area.

Figure 2

Headline inflation, underlying inflation and inflation expectations in the euro area

Headline inflation, underlying inflation and inflation expectations in the euro area

Evidence regarding inflation persistence is mixed, but points increasingly towards low persistence for the recent period

16A somewhat different question which arises in this respect is whether the price-stability-oriented conduct of monetary policy, apart from its clear impact on the average inflation rate during the period 1999-2008, has led to a decrease in inflation persistence in the euro area. This would be the case if the relevance of inflation expectations, as opposed to past inflation developments, increases for price-setting (switch from backward-looking indexation or adaptive expectations to forward-looking expectations) and if, in line with the rational expectations hypothesis underlying many theoretical models, credible monetary policy does actually succeed in isolating medium-term inflation expectations from current inflation developments. Below we review the empirical evidence regarding this particular question, which is somewhat mixed although it increasingly points towards low and/or decreasing persistence for the recent period.

17To illustrate the ongoing debate on this issue (rather than to provide a final answer on what has happened to inflation persistence in the euro area) we have estimated, for this article, the persistence of headline inflation and inflation excluding energy (core inflation, in the remainder of this section). In both cases, persistence is measured as the sum of all coefficients associated with lagged inflation in an autoregressive model. When this parameter is zero, inflation does not display any persistence at all and in fact behaves like white noise. When this parameter equals one, shocks to inflation have a permanent effect, implying that inflation no longer shows a tendency to converge towards a long-run value (unit root behaviour). Intermediate values, particularly when they are estimated to be significantly different from one, indicate that, notwithstanding some degree of persistence, inflation tends to converge to its long-run value, which in principle (and as discussed before) is pinned down by monetary policy. Lower values for the persistence parameter imply faster convergence. As the overriding aim is to assess whether inflation dynamics, and persistence in particular, have changed over time, both autoregressive models are estimated over a series of moving 48 quarters (12 years) windows. For headline inflation, the first estimate covers the period from 1970 I to 1982 II, while the last estimate refers to the period from 1996 III to 2008 III. Our data series for core inflation starts later than that for headline inflation, implying that, in this case, the first estimation period only starts in 1988 I (and runs to 1999 IV). Since persistence estimates obtained by ordinary least squares (OLS) tend to be downwardly biased, the median unbiased estimation developed by Hansen (1999) is used. Results are shown in the figure below in the form of the median unbiased persistence estimate and the upper and lower bounds of the 90% confidence interval. The null hypothesis of unit root behaviour is rejected when the upper bound of this interval is below one.

18For headline inflation, persistence is estimated as high for most of the windows considered, including those at the end of the sample. Most of the time, the estimate ranges between 0.8 and 1 and only for a few samples is it significantly different from one. Persistence does not display a tendency to decline for the more recent windows, which increasingly cover the EMU period. Interestingly, the results are different when we use a measure of core inflation. While for the earlier samples, persistence was again close to one, it starts to decrease as progressively more recent data are used in the estimate. For the windows ending in the first half of 2007, persistence is estimated at 0.6 and for quite a few windows ending in 2006 and 2007 the null hypothesis of a unit root for inflation can be rejected at the 90% level. However, for windows ending in late 2007 persistence increases again, before starting to fall at the very end of the sample. For the last window (period from 1996 III to 2002 III) median persistence is estimated at 0.7, which is considerably lower than at the beginning of our sample. A persistence parameter of 0.7 implies that the initial impact of a shock on inflation is reduced by half after 2 quarters, which suggests fairly fast convergence. However, as estimation uncertainty is large, persistence is not significantly different from 1 for the last window.

Figure 3

Inflation persistence, based on 12-year rolling windows [2]

Inflation persistence, based on 12-year rolling windows [2]

19The lack of any clear conclusion in our empirical exercise is also a characteristic of the relevant literature on this issue. Results tend to range from very high persistence (non-rejection of the unit root hypothesis) when very long samples are considered, to much lower, and in some cases even decreasing, persistence when account is taken of the fact that different inflation regimes may have been in place, mainly in terms of the average inflation rate. Various studies have in fact shown that failure to allow for a possible break in the level of inflation may lead to spuriously high persistence. [3] This was also one of the major findings of the Eurosystem Inflation Persistence Network, the results of which are summarised in the table below. Angeloni et al (2006) sum up these results as follows: “The bulk of the evidence (…) points to a moderate degree of reduced-form inflation persistence, once occasional shifts in the mean of inflation, which are most likely the result of shifts in the monetary policy regime, are accounted for. Within a stable monetary policy regime, the null of a unit root in inflation is generally rejected.”

Table 1

A selection of reduced-form estimates of inflation persistence (sum of autoregressive coefficients [4])

Euro area United States Inflation series used Sample and treatment of changes in the mean Simple estimates, assuming a single monetary policy regime (no mean changes) Altissimo, Mojon and Zaffaroni (2007) 0.93 CPI 1985: I - 2004: I Batini (2002) 0.74 HICP 1984: III - 2002: II Gadzinski and Orlandi (2004) [1.02; 1.04] [0.92; 1.03] GDP deflator, CPI, HICP and core inflation 1970: II - 2003: III O’Reilly and Whelan (2004) 0.96 HICP 1970: I - 2002: IV Robalo Marques (2004) 0.85 0.66 CPI 1982: I - 2002: IV Levin and Piger (2004) [0.65; 1.02] GDP deflator, CPI, PCE and core inflation 1984: I - 2003: IV Estimates, assuming changes in the monetary policy regime Dossche and Everaert (2005) 0.40 0.58 GDP deflator 1971: II - 2003: IV Time-varying inflation target Gadzinski and Orlandi (2004) [0.60; 0.90] [0.52; 0.80] GDP deflator, CPI, HICP and core inflation 1984: I - 2003: II Break in mean in 1993 (euro area) and in 1991 (US) Lünnemann and Mathä (2004) 0.40 HICP 1995: 1 - 2002: 12 O’Reilly and Whelan (2004) 0.90 HICP 1991: I - 2002: IV Robalo Marques (2004) 0.34 0.27 CPI 1986: II - 2002: IV HP-filtered series Levin and Piger (2004) [0.37; 0.89] GDP deflator, CPI, PCE and core inflation 1984: I - 2003: IV Break in mean in 1991 Source: table based on Altissimo et al. (2006)

A selection of reduced-form estimates of inflation persistence (sum of autoregressive coefficients [4])

20One major exception here is research by O’Reilly and Whelan (2005). Using rolling regressions, as we have done for this article, they find high inflation persistence even for their most recent window (1991-I to 2002-IV). This is also what we have found for that particular window and the picture of high persistence was furthermore confirmed for the later windows we have considered, although not for core inflation, a measure they do not consider. One plausible explanation for this divergence of the evidence based on rolling regressions could be that this approach is not particularly suited to dealing with shifts in the mean of the inflation process. Although estimation is based on fairly short windows, it could still be the case that mean shifts or more gradual changes in the mean have occurred within these windows. This is all the more so as the last sample considered by O’Reilly and Wheelan (2005) only uses three years of EMU data as opposed to nine years of data stemming from the pre-EMU period. Even our last sample still uses some data pertaining to the pre-EMU period. While considering even shorter windows could in principle help to overcome this problem, this will inevitably have the drawback that the reduction of the number of observations increases the width of the confidence interval even further and reduces the power of the unit root test.

21Some papers have estimated time-varying persistence and report a decrease in reduced-form persistence over the samples they consider. This is the case in Angeloni, Aucremanne and Ciccarelli (2006), as they find that persistence declined after the mid-1990s, but no longer during the EMU period. Although, in principle, this could be due to anticipated euro-effects during the run-up to the third and final stage of EMU, this interpretation is called into question by the fact that a similar decline occurred in the UK and the US too. This in itself does not invalidate the monetary policy explanation of the decrease in persistence completely, as it could still be the case that changes in domestic monetary policy occurred at more or less the same point in time in all these economies. More recently, Beechey and Österholm (2008) find that inflation persistence has fallen markedly since the third stage of EMU and inflation no longer exhibits unit root behaviour in an AR (1) model with constant mean and time-varying persistence parameter. They attribute this to the greater emphasis put on price stability.

22As emphasised by Angeloni et al (2006), the findings regarding moderate reduced-form persistence once occasional shifts in the mean are accounted for, also seem to be borne out by structural estimates of the degree of intrinsic inflation persistence, i.e. the part of persistence which comes on top of the persistence in the proximate drivers of inflation such as real marginal costs or the output gap (the latter part of persistence is often referred to as extrinsic persistence). Although there is generally a significant role for backward-looking aspects in price formation, the weight of this component drops when estimated over more recent samples or over stable monetary policy regimes. Benati (2008) finds that intrinsic persistence is extremely low or close to zero under EMU, as is his estimate of reduced-form persistence. Similar results are found for the UK, Canada, Sweden and New Zealand under inflation targeting, for the new Swiss monetary policy regime and for the Gold Standard. While this evidence shows that low inflation persistence is not an exclusive feature of EMU, it also suggests that creation of EMU clearly has played a role essentially by providing price-setters in the euro area with a stable and price-stability-oriented monetary policy regime.

Price-level convergence seems to have occurred prior to EMU

24The findings regarding EMU effects on price-level convergence are at best mixed in a lot of papers, and some prominent studies have found that the introduction of the single currency did not contribute to further market integration. Goldberg and Verboven (2004) report converging prices in the European car market between 1990 and 1992, but divergence thereafter. Engel and Rogers (2004) and Rogers (2007) find that price-level convergence occurred before 1994 and that no further progress was seen thereafter. Note, however, that Rogers (2007) also finds that price dispersion in the euro area is smaller than in the wider EU. Relative to the US, dispersion is of comparable size in the euro area for tradable goods and smaller for non tradables. In sharp contrast to the findings of these studies, Allington et al. (2005) do find a significant integrating effect for the euro.

25Divergences in results are perhaps due to the fact that databases of different nature are used. In the case of Engel and Rogers (2004) and Rogers (2007), data from the Economist Intelligence Unit (EIU) were used. This database has the advantage of containing genuine price-level observations for a representative set of individual goods, which are (fairly) comparable across countries. It nevertheless has the drawback of only containing data on capital cities, which could be rather unrepresentative of the whole economy. The database used by Allington et al. (2005) contains comparative price-level indices for a wide range of product groups, which Eurostat uses to construct purchasing power parity (PPP) series. While in principle more representative than the EIU dataset, the fact that indices are constructed for product groups implies that one no longer compares effectively observed price levels of individual goods. This leads to a downward aggregation bias in measured dispersion, as some price deviations with opposite signs would cancel each other out within a product group. However, the impact of the bias on the evolution of dispersion over time and on the comparison between the EMU countries and the EU, and hence on the results of Allington et al. (2005), is indeterminate.

No sign of structural changes in price-level stickiness as a result of EMU

26Based on unique sets of individual price records underlying the consumer price index of six European countries (Spain, Germany, France, Italy, Belgium and Austria), collected within the scope of the IPN, Angeloni, Aucremanne and Ciccarelli (2006) provide measures of price stickiness which can be considered as being representative for the euro area as a whole. One of their main findings is that all patterns in the period preceding and immediately following the start of stage three of EMU in 1999 are very stable. This suggests that the adoption of the single currency led to no change at all in pricing behaviour.

27Another finding of the IPN is that prices are quite sticky in the euro area compared to the US. The main IPN conclusions in this respect are summarised in the table below, which is inspired by Álvarez et al (2006). In any given month, only 15.1% of the prices of a sample of individual products representing the basket of goods and services making up the HICP are changed in the euro area (Dhyne et al., 2005), compared to 24.8% in the US (Bills and Klenow, 2004). As a result, the price of a product in the euro area is held constant for a period of 13 months, whereas that period is just under 7 months in the US. This relatively high degree of consumer price stickiness is confirmed by the results of specific price-setting surveys conducted by the IPN. The median life span of a price is close to 12 months in the euro area (Fabiani et al., 2006), whereas it is less than 9 months in the United States (Blinder et al., 1998). Producer prices are somewhat more flexible than consumer prices, as around 20% of them change every month on average. These microeconomic assessments of the degree of price-level stickiness are in general in line with the estimates obtained on the basis of macroeconomic analyses by Galí et al. (2001; 2003), who assessed the average life span of prices in the euro area at between 13.5 and just over 19 months, whereas in the US the figure was only 7 to 8.5 months.

28Dhyne et al. (2005) highlight various factors which can explain the difference in the frequency of price changes observed between the euro area and the United States. First, both the level and the variability of inflation were slightly higher in the US during the period analysed (January 1996 to December 2001). Second, small retailers still seem to hold a very important position compared to the super- and hypermarkets in the euro area, in comparison with the United States (Pilat, 1997). The results obtained in some euro area countries show that the super- and hypermarkets change their prices significantly more often than small retailers. Differences in statistical methodology are a third factor. In the majority of the euro area member states, the datasets used take no account of price changes associated with end-of-season sales, whereas those price changes are included in the American measurements. A fourth explanation might lie in the greater variability of wages and input prices in the United States. Conversely, the analysis showed that this difference was not due to divergences in the structure of consumption between the euro area and the United States, as services – which are the most rigid component of the HICP – account for a larger share of consumption in the United States than in Europe. If the consumption structure had been the same, the price change frequency differential between the two areas would have been even greater.

Table 2

Measures of the degree of price level stickiness

Price stickiness indicator Euro area United States CPI micro data1 Frequency (in% per month) 15.1 24.8 Duration (in months) 13.0 6.7 PPI micro data2 Frequency (in% per month) 20.0 10.8 - 13.3 Specific survey3 Duration (in months) 12.0 8.6 Macro – NKPC (GDP deflator) 4 Duration (in months) 13.5-19.2 7.2-8.4 1 Dhyne et al. (2005) for the euro area, Bils and Klenow (2004) for the United States. 2 Vermeulen et al. (2005) for the euro area, Nakamura and Steinsson (2008) for the United States. 3 Fabiani et al. (2005) for the euro area, Blinder et al. (1998) for the United States. 4 Galí et al. (1999, 2001, 2003). Source: table based on Alvarez (2006).

Measures of the degree of price level stickiness

29More recent work for the US (Nakamura and Steinsson, 2008) points to a new factor which has the potential to provide an additional explanation for the gap between the euro area and the United States, namely the importance of sales prices (temporary promotions) in CPI data. In their paper, Nakamura and Steinsson find that the median frequency of non-sales price changes is roughly half of what it is including sales in the United States (9-12% per month versus 19-20% for identical items; 11-13% versus 21-22% including product substitutions). On the contrary, results from Baudry et al., 2004, and Baumgartner et al., 2004 for, respectively, France and Austria show that temporary promotions tend to increase the overall frequency of price changes by only 3% in European countries. This recent evidence on the US also suggests a higher degree of stickiness for producer prices than in the euro area.