SEMINAR. 12. Is there a discrepancy between measured and perceived inflation in the euro area countries since the euro cash changeover?

Transcription

1 SEMINAR Inflation Measures: Too High - Too Low - Internationally Comparable? Paris, - June. Is there a discrepancy between measured and perceived inflation in the euro area countries since the euro cash changeover? L. Aucremanne, M. Collin and E. Dhyne (Central Bank of Belgium)

2 . Is there a discrepancy between measured and perceived inflation in the euro area since the euro cash changeover? Luc Aucremanne ( * ) Marianne Collin (**) First Draft prepared for the OECD seminar on CPI, June, Abstract: Using unit root tests for the difference between re scaled inflation perceptions, as measured by the EC survey with consumers, and actual inflation for euro area countries, we find that perceptions track measured inflation quite well during the period up to December. However, for the whole sample period considered (up to April ) the null hypothesis of a unit root can no longer be rejected, suggesting that the relationship between perceived and measured inflation breaks. In a control panel of non euro area members no such break is found, implying that the cash changeover was an important triggering factor of the perception gap in the euro area. Results are more or less the same when national CPI inflation is used instead of HICP inflation. Therefore, the misperception is not specifically related to the HICP. We do not find econometric support for the hypothesis that inflation perceptions are based on a restricted sample of frequently purchased goods and services. In particular, perceptions diverged substantially from such an alternative inflation measure in the period before. Finally, the fraction of the perception gap that persisted until the end of our sample (April ) displays some positive correlation, across euro area countries, with recent developments in real house prices. Further investigation is however needed to assess whether the perception gap is indeed related to the non inclusion of owner occupied housing services in the HICP. Key words: inflation, perceived inflation, unit root tests, panel unit root tests JEL code: C, C, D, E (*) (**) NBB, Research Department, (e mail: luc.aucremanne@nbb.be). NBB, Research Department, (e mail:marianne.collin@nbb.be). Inflation Measures: Too High Too Low Internationally Comparable? Paris, June

3 .. INTRODUCTION In many papers on recent euro area inflation developments it is shown that the inflation perceptions of consumers diverge substantially from measured inflation since the euro cash changeover in January. This phenomenon is often illustrated by the recently grown gap between perceived inflation, as measured by the EC survey with consumers, on the one hand, and HICP inflation, on the other. The deviation of consumers' inflation perception from actual HICP inflation started in, after the changeover and reached a peak around January in each of the euro area economies. Since then, a downward adjustment of perceived inflation has taken place in most countries. There is, however, a substantial degree of heterogeneity in the way the gap is closing. Indeed, at the current juncture, we still observe a relatively large discrepancy between perceived inflation and HICP inflation in most of the countries, while in Germany the gap seems to have disappeared. In many cases this perception gap is discussed on the basis of a purely graphical analysis, in which both series actual and perceived inflation are plotted together, using a specific scale for each series. The aim of this paper is to address this issue in a more formal, econometric setting and to see whether the inference made from a graphical comparison of the quantitative and the qualitative inflation measures makes also sense in such a setting. The paper intends in the first instance to verify whether perceived inflation tracks measured inflation in a reasonable manner in the pre euro period. It therefore transforms the qualitative measure of inflation perceptions and tests whether the difference between this re scaled perception measure and actual inflation is stationary over the period up to December. Unit root tests on this difference are performed for the euro area as a whole and for the individual euro area countries. Moreover the data for the individual countries are pooled and a panel unit root test is performed, in order to enhance the power of the test. Rejection of the null of a unit root would imply that perceived and measured inflation indeed co moved in the pre euro period and that differences between both measures tended to be short lived. The test also yields a measure of the persistence of differences between both inflation measures. See, for instance, European Central Bank (a, b, and a), European Commission (), National Bank of Belgium (), Cornille () and Deutsche Bundesbank ()

4 . An identical test is then performed for the entire period, including the period after the introduction of the euro. Comparing the results of both tests allows us to check whether the inclusion of the more recent period affects the degree of persistence of the difference between perceived and measured inflation in a significant manner. In order to assess the role of the introduction of the euro, the same tests are performed, for identical subperiods, on a control panel of European countries outside the euro area (Denmark, Sweden and the UK). Subsequently, we test alternative explanations of the gap, for instance the possibility that inflation perceptions are based on national CPI figures, rather than on the HICP measure or the thesis according to which inflation perceptions are based on a subset of frequently purchased goods and services. Finally, we also test the hypothesis whether the gap is related to the non inclusion of owner occupied housing services in the HICP. This is (indirectly) done by comparing different gap measures at the country level with the recent evolution of property prices in real terms. The paper is organized as follows. In section, we present the data set. Section describes the methodology used in this paper. In section, we discuss the results of our unit root tests and interpret them against the background of the alternative explanations for the perception gap. Finally, section concludes.. DATA Our data set consists of three types of inflation measures. For all series we have considered data up to April, while the starting date depends both on the series considered and on the country considered. The three inflation measures are available for individual euro area countries, for the euro area as a whole, and for Denmark, Sweden and the UK. As to the individual euro area countries, Luxembourg is not in our analysis, as no data on inflation perceptions were available before December. Denmark, Sweden and the UK are used as a control panel, as in these countries inflation perceptions have not been affected by the euro cash changeover. The first type of inflation measures are based on the harmonized indices of consumer prices (HICP), published by Eurostat. HICP data start in January 99, implying that inflation rates are available from January 99 onwards. Note that for some countries HICP data are available before 99. We have, however, opted for an identical starting

5 . date of the series for all of the countries considered. The HICP data are our benchmark statistic for observed (or actual) inflation. The second type of inflation measures are the national CPIs. We have considered CPI data as an alternative for HICP data, for two reasons. First, as HICP data are not always well known by the general public, it could be the case that inflation perceptions are based on the evolution of the CPI, rather than on the evolution of the HICP. In that event, observing a gap between HICP inflation and perceived inflation does not necessarily imply that a similar gap exists for CPI inflation. Second, CPI data are available for a longer time span than HICP data. This allows us to check whether inflation perceptions track measured inflation over a longer time period. For our analysis, we have considered CPI data from January 9 onwards, implying that our CPI inflation series start in January 9. These data are taken from the IFS of the IMF. The third type of inflation measures are the qualitative data on inflation perceptions, as measured by the consumer survey conducted by the European Commission (EC). Each month, the EC questions around, consumers in the euro area about their perception of inflation. The particular question is: "How do you think that consumer prices have developed over the last months?" Note that the particular question refers to price developments over the last twelve months. This reference period of one year, explains why we have computed yearly inflation rates for the HICP and the CPI data. There are six possible answers to the question on inflation perceptions: () risen a lot, () risen moderately, () risen slightly, () stayed about the same, () fallen, () do not know. The results of these surveys are summarized as a "balance statistic", calculated as the difference between the weighted proportion of respondents stating that consumer prices have risen and the weighted proportion of consumers stating that consumer prices have fallen or stayed about the same. This balance statistic is a qualitative measure of inflation perceptions. Therefore, it provides only information on the directional change of perceived inflation and, in principle, it cannot provide a quantitative measure of inflation. In the euro area, these data are available from January 9 onwards, for Belgium, France, Germany, Greece, Italy, The Netherlands, Portugal and Spain. These countries form the base group in our analysis. We consider also an alternative base group that excludes Greece. Indeed, this country seems to be characterized by a particularly large perception gap, which is difficult to interpret, given the substantial change in the monetary policy regime and the accompanying shift towards price stability in Greece during the As a matter of fact, CPI inflation is available before 9.. However, we have not considered the data before 9., as our data series on perceived inflation start only in 9..

6 . sample period. Qualitative survey data are only available from January 99 onwards for Austria, Finland and Ireland. Together with the countries in the base group, these countries form the euro area group, as no data on inflation perceptions are available for Luxembourg before December. Here too, we consider an alternative euro area group excluding Greece. In our control panel, data are available since January 9 for Denmark and the UK and since January 99 for Sweden. These differences in terms of data availability, both across series and across countries, will have an important impact on the practical implementation of our econometric analysis. All available data series are shown in Figures and, where the qualitative survey data are plotted in their original form, together with HICP inflation data for the period starting in 99 (Figure ) and CPI inflation data for the period starting in 9 (Figure ).. METHODOLOGY Notwithstanding the important difference in the type of statistics used, we will compare in this study quantitative inflation measures (HICP and CPI) directly with a qualitative measure of perceived inflation. Obviously, this comes as an important caveat to our analysis. However, this is exactly what most of the other studies on this issue have done, often in a purely graphical analysis. Our approach can be seen as a very pragmatic one: we just test whether the inference made from a graphical comparison of the quantitative and the qualitative inflation measures makes sense in a more formal econometric setting. This does not mean, however, that we use the raw data on perceived inflation. For the purpose of this study, we have transformed the survey data by dividing the original series by the ratio of the standard deviation of the survey data, on one hand, and the standard deviation of inflation, on the other. P Bit π it = () se( Bit ) se( π ) it P where π it represents our (transformed) measure of perceived inflation, Bit denotes the original balance statistic from the survey, and se( B it ) and se( π it ) stand respectively for the standard deviation of the balance statistic and the officially measured inflation (HICP or national CPI).

7 . The transformed survey data are plotted in Figures and, respectively together with HICP inflation and with CPI inflation. This data transformation has several advantages. First of all, it controls for differences, across countries, in the sensitivity of inflation perceptions to actual inflation developments. An increase in the balance statistic of a given magnitude, does not necessarily match similar movements in inflation (in case of such differences in sensitivity). Second, it implies that the transformed survey data have the same standard deviation as observed inflation. As a result, they are more directly comparable to actual inflation. Third, this data transformation is very similar to what is done implicitly in a purely graphical exercise, whereby often two different scales are used for actual and perceived inflation. Fourth, as both concepts have the same standard deviation, the difference between the re scaled survey data and actual inflation can provide a meaningful measure of the perception gap, under certain conditions. In particular, in case perceived inflation tracks measured inflation in a reasonable manner, the difference between both inflation measures ( D = π π P i, t it it ) should be stationary around a constant. The constant itself is irrelevant, as the absolute level of the balance statistic is not meaningful. The latter idea is central in our empirical strategy. We will indeed test whether the difference between the re scaled series for perceived inflation and actual inflation is stationary. We will do this test, in the first instance, for the period up to the date where the gap seems to appear. This is for the period 99.., when HICP data are used as actual inflation measure. Rejection of the null of a unit root would imply that perceived and measured inflation indeed co moved in the pre euro period and that differences between both measures tended to be short lived. The test also yields a measure of the persistence of differences between both inflation measures. In the graphical analyses, the existence of a simple and stable relation between perceived and actual inflation for the period up to is assumed implicitly and verified by visual inspection only. Our empirical strategy allows us to test this assumption formally. Such a formal test seemed justified to us, as from a visual inspection of the graphs in Figures to it is not always clear whether perceived inflation tracks actual inflation at the level of the individual countries. Finding a stationary difference during this reference period is a prerequisite for addressing the question whether something particular happened in the following period. Indeed, in the case the difference is not stationary, there was no stable, simple relation between rescaled perceived inflation and actual inflation in the past and, hence, the issue of instability in the more recent sub period is irrelevant.

8 7. Thereafter, we will perform an identical test for the entire period (including the period for which there seems to be a gap). Comparing the results of both tests allows us to check whether the inclusion of the more recent period affects the amplitude and the degree of persistence of the difference between perceived and measured inflation in a significant manner. In case a severe and persistent gap has occurred during the more recent period (i.e. starting in ), the difference between the transformed series for perceived inflation and actual inflation would no longer be stationary over the entire sample (99.., in the case HICP data are used as actual inflation measure). As we test for stability, it is clear that for the entire sample an identical data transformation has to be made as for the shorter sample. This implies that the ratio of the two standard deviations continues to be based on the shorter sample (99..). Using a ratio based on the entire period, would in contrast accommodate to some extent the possible break and therefore induce a bias towards finding stability. These tests are performed for the euro area as a whole and for the individual countries. In particular, the following traditional augmented Dickey and Fuller (ADF) model (see Dickey and Fuller (9)) is estimated: D pi i, t = ci + ρ i Di, t + βi, j Di, t j + ε i, t j= where i indexes the geographical areas considered, () ρ i is the sum of the autoregressive coefficients in an AR model of order p i, D it is our gap measure defined above and it stands for the residuals. The optimal lag length or, equivalently, the order p i of the AR model is obtained on the basis of the SIC criterion. Under the null hypothesis, D it is assumed to have a unit root, i.e. the data generating process is said to be non stationary. Under the alternative hypothesis, the time series is stationary. In other words, H : i = and H : < i < Analyzing the individual countries (instead of only the euro area as a whole) seemed relevant for two reasons. First of all, from a conceptual point of view, the euro area is merely a statistical construction for the period prior to 999. It therefore makes more sense to perform our test on the individual countries, as part of our sample is before the start of EMU. Second, it is well known in the econometric literature that unit root tests have low power in short samples. This problem is potentially severe for our data, as our benchmark period only covers years of monthly observations. Standard unit root tests have therefore difficulties in distinguishing non stationary time series from stationary. In

9 . particular, the unit root test tends to fail rejecting the null hypothesis of a unit root even if it is not true. In order to avoid this problem, we use the individual country data to constitute a balanced panel for which we will perform panel unit root tests. These tests have more power, as they exploit the extra information contained in the pooled cross section of the time series. Recently, multiple panel unit root tests have been developed in the literature. In this empirical study, we present results for three different tests in order to check the robustness of our findings: Levin, Lin, and Chu (), Breitung () and Im, Pesaran and Shin (997). The tests developed by Levin, Lin, and Chu (LLC) and by Breitung () assume that the persistence parameter is common for all individuals (countries in our case) in the cross section. In contrast, the test developed by Im, Pesaran and Shin (997), IPS hereafter, allows for the persistence parameter to vary freely across individuals in the cross section. Imposing a common persistence parameter (as is the case in the first two tests) can indeed spuriously lead to the non rejection of a unit root, in case of stationary processes with heterogeneity in the persistence parameter. As mentioned above, we defined in the panel data analysis different groups of countries due to the heterogeneity in the starting date of the individual times series. For the panel unit root tests on the basis of the HICP and the CPI over the shorter period 99.., we identified five different groups: the base group, which includes Belgium, France, Germany, Greece, Italy, the Netherlands, Portugal and Spain; the base group, excluding Greece; the euro area group, which includes all euro area members, except Luxembourg; the euro area group, excluding Greece; the control group, which includes Denmark, Sweden and the UK. For the longest period, namely 9.., data availability urges us to perform the panel unit root tests on the basis of the national CPI only for the two alternative base groups and the control group, which excludes, in this particular case, Sweden. Finally, for the purpose of this empirical study, we have built a new index, which includes only frequently purchased goods or services, i.e. non durable goods and daily consumer purchases, such as food, beverages, tobacco, transport services, fuel, postal services, hotels, cafes and hairdressing. The composition of this restricted basket is based on a similar index developed by the ECB (). The items included in this index represent

10 9. around % of the total HICP basket. We refer to this index as the out of pocket index. As this index is based on (detailed) HICP data, the corresponding inflation rates are only available for the shorter period Therefore, we perform for this index panel unit root tests for the same groups as for the HICP, except for the control group, which excludes Sweden, due to the lack of the detailed data series needed for the construction of this alternative index. The inflation rates based on this "out of pocket" index are shown in Figure. They are compared with re scaled survey data. Evidently this transformation, described by formula (), makes now use of the standard deviation for the alternative "out of pocket" inflation measure. This implies that the re scaled perception measures differ to some extent from those already presented in Figures and.. RESULTS In this section, we present our results of the unit root tests for the individual countries and for the different panels considered. We will structure this presentation of results along the lines of a series of questions. The first questions address the issue of whether there is indeed a perception gap and the subsequent questions address the different factors which are frequently mentioned as being responsible for the gap. For each factor, we investigate empirically to what extent it might have contributed to the perception gap. QUESTION : DO INFLATION PERCEPTIONS MEASURED BY THE EC SURVEY WITH CONSUMERS TRACK HICP INFLATION IN THE PERIOD UP TO DECEMBER? Table presents the results of the unit root tests on the difference between re scaled inflation perceptions and HICP inflation, for the individual countries in our dataset. Table presents similar results for the panel unit root tests. The left panels of the Tables and present results for the period 99.., while the right panels present results for the entire sample (99..). We concentrate first on the left panels, in order to address our first question. For the euro area as whole, we are not able to reject the unit root hypothesis for the difference between re scaled perceived inflation and HICP inflation during the period at the % significance level. The estimated persistence parameter (.9) is indeed quite high and relatively close to unity. At the level of the individual countries, the results are more mixed for the pre period, since we are able to reject the unit root hypothesis for countries (Belgium, Germany, the Netherlands and Portugal), while for the other countries (including large countries such as France, Italy and

11 . Spain) the null of a unit root could not be rejected. Based on this evidence, one would rather reject than accept the existence of relation between perceived and actual inflation, even for the period prior to the euro cash changeover. However, in the economic literature on unit root tests it is well known that these tests have low power in short samples and in particular, they tend to fail rejecting the unit root hypothesis even if it is false. Panel unit root tests constitute a solution to this problem, as they increase the power of the tests by exploiting the extra information in the crosssection. Table presents the results of these panel unit root tests obtained on the basis of different methodologies available in the literature and for different panels of countries. The choices we made in terms of the different panels considered were motivated in sections and. The results for the panel unit root tests, presented in Table, are unambiguous. The unit root hypothesis is firmly rejected for the pre euro period at the % significance level for all the different groups considered; the value of the t statistic is always higher in absolute value than the critical values for all the available tests. This evidence suggests that, on average, consumers have relatively well assessed the level of inflation before the cash changeover and that differences between perceived and actual inflation were relatively short lived. Overall, we conclude from this that perceived inflation, as measured by the EC survey with consumers, tracked HICP inflation reasonably well during the period Note however that it is was not that straightforward to verify this relation. First of all, it was absolutely necessary to re scale the original balance statistic, along the lines explained in Section. Indeed, without this transformation the hypothesis of a unit root could not be rejected for the euro area as whole, for out of the individual countries and for all of the panels considered. Tables bis and bis give the results of these tests using the original balance statistic of the EC. Second, given the short sample considered, it was absolutely necessary to rely on a panel data approach in order to obtain an unambiguous positive answer to our first question. QUESTION : DO INFLATION PERCEPTIONS MEASURED BY THE EC SURVEY WITH CONSUMERS CONTINUE TO TRACK HICP INFLATION AFTER? In stark contrast to the results for the period up to December, when considering also the period after the introduction of the euro banknotes and coins (i.e. the whole sample 99..) the null hypothesis of a unit root can never be rejected for the individual

12 . countries. On top of that, a similar result is found for each of the panels considered, the t statistic of these tests being, in absolute value, always below the critical values. Here, the panel data analysis corroborates the results obtained on the basis of the unit root tests for the individual countries. This is strong evidence in favor of a break in the relation between perceived and measured inflation. QUESTION : IS THE PERCEPTION GAP PERSISTENT? Given our methodology described in Section, we have now a straightforward measure of the perception gap. Indeed, we simply consider the difference D it, corrected for the unconditional constant in the stochastic process described in equation (), as a quantitative measure of the perception gap. In other words, we define the perception gap as: GAP = D c i i, t i, t () ( ρi ) Obviously, we use here the unconditional mean estimated over the shorter sample (99..). Given our data transformation, our measure for the perception gap has the same units as inflation and therefore its magnitude can be interpreted in terms of percentage points of annual inflation. The graphs in Figure are constructed accordingly, i.e. the right hand scale has identical units as the left hand scale and it has been corrected for the (irrelevant) unconditional constant of the stochastic process. Therefore, it is completely identical and not plotted separately. As a result, the distance between the two curves measures the magnitude of the perception gap. While in all countries relatively small and short lived perception gaps are observed in the period prior to, the gap increases sharply and in a relatively synchronized fashion in the period surrounding the cash changeover in January. In most of the countries the gap further increases well into and thereafter it stabilizes or, in some cases, it is reduced. Notwithstanding the overall movement towards smaller gaps in the period, in many cases a non negligible gap persists even until the end of our sample (April ). Germany is the only country where the gap is completely closed. For the euro area as whole, our gap measure reached a maximum of approximately percentage points of

13 . annual inflation and, at the current juncture, it still exceeds percentage point of annual inflation. Even this very persistent part of the gap is still quite substantial. Overall, these results confirm that a break between inflation perceptions and officially measured inflation occurred, giving rise to a perception gap which is pretty persistent. Our econometric analysis therefore tends to confirm the inference often made from purely graphical inspection of the time series. In the next set of questions, we will assess empirically the relevance of some of the factors which have been put forward to explain this phenomenon. QUESTION : IS THE EURO CASH CHANGEOVER RESPONSIBLE FOR THE BREAK IN AND FOR THE PERCEPTION GAP? While the timing of the break is indeed compatible with this hypothesis, we also checked this hypothesis by performing similar tests for a control panel of three non euro area countries for which similar data are available, namely Denmark, Sweden and the UK. In principle, in these countries inflation perceptions should not have been affected by the introduction of the euro in. As shown in Table, results for this control panel are strikingly different from what we found for the euro area. At the level of the individual countries, the unit root hypothesis is rejected for Sweden at the % significance level (and for Denmark at the % significance level) during the pre euro period. For the U.K, however, the unit root cannot be rejected for the short sample period. For the whole sample period, results are nearly unaffected by the inclusion of the post period. The estimated persistence parameters are indeed very similar to those for the shorter sample, while for the euro area countries we always found an increase in the persistence parameters. In the case of Denmark, the unit root hypothesis is now even rejected at the % significance level. Table reports results for the panel unit root tests. Most of the tests ( out of ) reject the null hypothesis of a unit root for the pre period and, interestingly, also for the entire sample. Overall, this suggests that in these non euro area countries perceived inflation tracks actual inflation relatively well, both in the pre and the post period. In contrast to the euro area countries, we do not detect any asymmetry in the way consumers perceive inflation after. We interpret these results as strong evidence in favor of the euro cash changeover as the main explanatory factor for the discrepancy between perceived inflation and officially measured HICP inflation in the euro area countries.

14 . This conclusion is corroborated by the opinion survey conducted by the EC on the general public and the euro (), which revealed that a vast majority of the respondents are convinced that the conversion of prices into euro has been to the detriment of consumers. In particular, Figure shows a positive relationship across countries between, on the one hand, the scope of the perception gap (averaged over the period ) and, on the other hand, the conviction that the conversion of prices into euro has been to the detriment of consumers. Del Giovane and Sabbatini () found a similar positive relationship. However, the link is somewhat stronger in their analysis, although they used the balance statistic in its original form. It is worth mentioning that these opinion surveys and the evidence in favor of the euro as the main factor behind the gap is in contrast with most studies conducted in the euro area, which have pointed out that the introduction of the euro had a relatively small impact on official HICP inflation. Moreover, there is no reason at all to think that the HICP is not capturing the impact of the cash changeover accurately. Many studies have indeed shown that, at the micro level, the transition is well reflected in the official price statistics (see for instance Cornille ()). The observed gap is therefore due to misperceptions and not to inaccurate measurement. While it is relatively easy to understand that the euro cash changeover confused consumers and led temporarily to inflation misperceptions, it is more difficult to understand why the perception gap is so persistent. Therefore, we investigate below three other factors, which might have contributed to the persistence in the gap. They were selected on the basis of a combination of the following criteria: they are all, to some extent, related to particular aspects of inflation measurement and they are all relatively easy to test in our methodological framework described in Section. We subsequently address the questions whether the perception gap is due to (i) a lack of credibility for the HICP, (ii) the fact that consumers form expectations on a limited subset of frequently purchased goods and services, (iii) the non inclusion of owner occupied housing in the HICP. Note however that our list of factors is not exhaustive. Indeed, several other explanations have been put forward as well. A first argument frequently mentioned is the fact that consumers might have incorrectly interpreted their loss of purchasing power as being The exact question asked by the EC in the Flash Eurobarometer () is the following: "Did you personally notice that, in your country when converted into euro, prices have been: rather in favour of consumers; rather to the detriment of consumers; or one way or another the rises and falls balanced out?". Del Giovane and Sabbatini show in their working paper a strong positive relationship (R²=.) between the change in inflation perceptions based on the original data of the EC, on the one hand, and on the other hand, the conviction that the conversion of prices into euro has been to the detriment of consumers. See, among others, Cornille () for a specific analysis of the inflationary impact of the cash changeover in Belgium.

15 . caused by higher inflation rather than by the general slowdown in the economic activity. The large media coverage of the euro cash changeover and in particular the enhanced consumer resistance to price increases may also have contributed to the gap. Del Giovane and Sabbatini () found, for instance, evidence in favor of the first argument in the case of Italy, while they do not find any convincing support for the second. Recently, studies have highlighted the role played by psychological factors in explaining the gap and in particular, its long lasting effect. Surveys 7 have indeed found that consumers' perceptions are somewhat biased, as a result of a priori expectations of price increases. Therefore, they tend to overestimate price increases, whilst if prices do decrease they are perceived as prices having remained stable. Another element is the difficulty that consumers have in handling with the new currency, implying that the national currencies remain the reference point for many consumers. This factor can indeed produce persistence in the gap, as actual prices continue to be compared with the last prices in the national currency, observed before the introduction of the euro. As time progresses, these reference points are more and more outdated. At the current juncture, we have not examined the empirical relevance of these alternative explanations. QUESTION : DO CONSUMERS FORM THEIR INFLATION PERCEPTIONS ON THE BASIS OF THE MOST POPULAR INFLATION MEASURE, THE NATIONAL CPI, RATHER THAN ON THE BASIS OF THE HICP? It has often been argued that national CPIs are still the primary reference price indices at the national level and that the HICPs are ranked as secondary indicators. This attitude could in part be due to the fact that the National Statistical Institutes (NSIs), in their external communication, tend to give more priority to the national consumer price indices than to the corresponding HICPs. In addition, in some countries the national CPIs are still used as reference for indexation mechanisms (wages, contracts,..). Finally, as regards the particular measure of perceived inflation used, it should be noted that the EC survey does not make reference to a particular consumer price index when asking respondents to assess the evolution of prices (in general) over the last months. As a result, it is possible that consumers form their inflation perceptions on the basis of their experiences with the national CPIs, rather than the HICPs. Figure shows, for each individual country, the evolution of re scaled inflation perceptions, on one hand, and of national CPI inflation on the other. On the basis of these charts, we observe a relatively close relationship between perceived inflation and CPI 7 See for instance Traut Mattausch et al. () for Germany. See Mastrobuoni ()

16 . inflation for the pre period, but thereafter a relatively important and persistent deviation between these two inflation measures is observed. This suggests that also the link between perceived inflation and national CPI inflation breaks down in. Formal tests confirm this observation. Tables and report, for the period 99.., the results of unit root tests for the difference between re scaled inflation perceptions and national CPI inflation, respectively for the individual countries and for the different panels considered. Both categories of tests are completely in line with the previous results based on the HICP. First, at the level of the individual countries, the unit root hypothesis can never be rejected over the whole sample period (99..), while for the pre euro period the evidence is less clear cut, as for countries the null hypothesis of a unit root can be rejected, while it cannot for the remaining countries. Second, panel unit root tests do provide convincing evidence of a break in. Indeed, the unit root hypothesis is, in general, rejected for the pre period. Over the whole sample period, in contrast, we are never able to reject the null hypothesis of a unit root. Besides these tests for the euro area countries, we analyze whether the same kind of discrepancy is observed for the non euro area countries. Again, very different results are found, as we do not observe any significant difference between the pre and post periods. The unit root is, in general, rejected for both time periods, suggesting that perceived inflation in these non euro area countries tracks national CPI inflation relatively well over the whole sample. In sum, our findings on the perception gap in terms of the HICP are to a large extent also valid for the national CPIs. Inflation misperception therefore seems to be a relatively widespread phenomenon and not just an issue characterising the HICP. From the point of view of assessing the accuracy and the credibility of the HICP specifically, this is a rather favorable finding. QUESTION : ARE THESE FINDINGS ROBUST FOR AN EXTENSION OF THE SAMPLE PERIOD? It might be argued that the sample period considered so far is too short and/or somewhat random and therefore does not allow drawing strong conclusions. In particular, it could be the case that the actual discrepancies are not exceptional by historical standards, once the sample period is extended. For the link between inflation perceptions and national CPI inflation such an extension is possible, as data are available for a relatively long time period, namely 9...

17 . Tables and show that the results are very similar to what we have found for the shorter time period, not only at the level of the individual countries, but also for the panel data. For the individual countries, the unit root hypothesis cannot be ruled out for the whole sample, except for Belgium. However, for the pre euro period we are able to reject the null hypothesis at the % significance level for out of countries and at the % significance level for all the countries available. Rejection of the null of a unit root at the level of the individual countries during the pre period turns out to be easier than in the case of the short sample, which is normal as the power of the tests is enhanced by the extension of the sample period. Therefore, there is less of an incentive to use the panel unit root tests in this case. Their results remain, however, interesting. The unit root hypothesis is strongly rejected for the pre euro period and, in contrast, for the whole sample period the null hypothesis cannot be rejected at all (Table ). Interestingly, we again find relative stability in the relation between perceived inflation and measured CPI inflation for the control group, both at the level of the individual countries and at the level of the panel. Our results obtained are remarkably robust for the extension of the sample period. QUESTION 7: ARE INFLATION PERCEPTIONS BASED ON A SUBSET OF FREQUENTLY PURCHASED GOODS AND SERVICES FOR WHICH PRICE INCREASES HAVE BEEN MUCH MORE PRONOUNCED DURING RECENT YEARS? It is often said that consumers form their inflation perceptions on the basis of a restricted basket of goods and services, which are frequently purchased, such as bread, meat, restaurant and bar services,... This argument has been mentioned as the most important factor behind the asymmetry in perceived inflation and officially measured inflation since January, because the prices of this type of goods have sharply increased in most euro area countries, more or less at the time of the cash changeover. Indeed, a succession of adverse temporary shocks such as the animal disease (BSE) in, cold weather in January, or the dry summer in has exerted pronounced upward pressures on inflation. Moreover, since oil prices have been nearly permanently under upward pressure. Petrol, food (in particular meat and vegetables) and restaurants are products for which relatively high price increases have been observed. As a result, it is possible that consumers have wrongly associated this clustering of temporary shocks to HICP inflation with the cash changeover. According to some studies, this factor was relatively important in explaining the perception gap, in particular in Italy and in Spain 9. 9 See Del Giovane and Sabbatini () and Bank of Spain ()

18 7. We test this hypothesis formally by re constructing an index, called by the ECB () the "out of pocket" index, which includes non durable goods and daily consumer purchases (i.e. food, beverages, tobacco, transport services, fuel, postal services, hotels, cafés and hairdressing). The items included in this index represent around % of the total HICP basket. Figure illustrates the evolution of "out of pocket" inflation and the accordingly re scaled inflation perceptions. For these charts, the conclusions are twofold. First of all, consumers do not seem to base their perceptions exclusively on a sub set of frequently purchased goods, as the link between this new price index and perceived inflation does not appear to be very strong over the whole sample period. Unit root tests on individual data and on pooled time series confirm this conclusion, as the unit root can never be rejected, even not for the pre period (Table 7 and ). The only exception is Portugal. It is in this respect also interesting to see that, while the unit root hypothesis can be rejected for the control panel for the short sample, it can not for the longer sample. This is further evidence in favour of structural i.e. unrelated to the euro introduction instability in the relation between perceived inflation and "out of the pocket" inflation. Second, the increase of the "out of pocket" price index has been somewhat above the HICP inflation over the period.. in some countries such as Italy, Portugal, Greece and the Netherlands, and as a consequence, the gap between the "out of pocket" price index and the perceptions of inflation appears to be lower than on the basis of the entire HICP in these countries. Overall, although this factor may have played a role in some countries in explaining the discrepancy between the inflation perceptions and actual inflation at the time of the introduction of the euro, we do not think it is an important element. We indeed found that there is a lot of structural instability in the relation between perceived inflation and "out of the pocket" inflation, even in the period prior to the introduction of the euro. A more or less similar conclusion is drawn in ECB (). QUESTION : HAS THE NON INCLUSION OF OWNER OCCUPIED HOUSING SERVICES CONTRIBUTED TO THE PERCEPTION GAP? An important shortcoming of the HICP at the current juncture is the non inclusion of owner occupied housing. However, the purchase of houses constitutes by far the most important transaction of households. In recent years, house prices have moreover substantially risen in a large number of euro area countries and this increase has been in general well

19 . above HICP inflation. In the euro area as a whole, residential property prices have increased by 7.% in, after a growth of 7% in (ECB, b). These sharp increases in house prices might have had an important impact on the wealth of the economic agents, given that housing is the largest asset in their portfolio. Furthermore, this widespread phenomenon on the housing market has received extensive media coverage and, as a result, consumers might have been very sensitive to this price acceleration. On top of that, it is reasonable to think that respondents to the EC survey and households in general are not conscious of the fact that dwelling prices are not (yet) included in the HICP. House price developments might therefore have contributed to some extent to the scope and the persistence of the gap in most euro area countries. Figure 7 presents different charts showing the link between a set of alternative measures of the perception gap and the developments in real house prices. The first and second graph present the relationship between the average perception gap since the cash changeover ( ) and average real house price increases over the time period and respectively. From these charts, it is not very clear whether there is any relationship between the perception gap and the "boom" of the housing market. However, the following graphs, aimed at linking the persistent part of the gap to real house price developments averaged over different time periods, are somewhat more conclusive. Pronounced real house price developments tend to be associated with relatively larger gaps in between perceived inflation and HICP inflation, except for Greece and Portugal where house price increases have been relatively moderate, while the perception gap is pronounced. Finding a better correlation with house prices for the persistent part of the gap than for the average gap seems logic, as the latter gap concept is strongly affected by the misperceptions induced by the cash changeover, which is in principle orthogonal to house price developments. Correlations between the part of the gap that persisted into and real house price increases are, however, relatively low (about.) and turn out to be not statistically significant, even when excluding Greece. Overall, this evidence suggests that house prices could have, to a relatively limited extent, contributed to the persistent discrepancy between perceived and actual HICP inflation. It is however hard to draw strong conclusions on the basis of this relatively simple empirical exercise. Further research is therefore needed and probably interest rate developments and differences across countries in the relative importance of owner occupied housing should be brought into the picture. This is measured by either the average gap for the first months of or by the gap for the last month, April.

20 9.. CONCLUSIONS While there is clearly no doubt at all about the accuracy of the official inflation measures during the recent period, since euro area consumers tend systematically to perceive that inflation is high, while in reality it is relatively low. Apparently a perception gap has grown, starting in and lasting until today. In this paper we test econometrically whether there is indeed such a perception gap and we also examine the empirical relevance of some of the factors that are frequently put forward as possible explanations. We confirm the fact that (re scaled) perceived inflation tracks measured HICP inflation relatively well in the pre euro period (i.e. until December ), both for the euro area countries and for the other EU countries. In earlier analyses, this relation was mainly verified on graphical grounds only. There is moreover econometric evidence that this relation breaks in in the euro area, giving rise to a gap between perceived and actual inflation. In a control panel of non euro area countries (Denmark, Sweden and the UK) no break is found in or in the following years. The euro cash changeover therefore seems the triggering factor of the perception gap in the euro area. Earlier graphical evidence on a break is confirmed econometrically. We were able to construct a quantified gap measure and, according to this measure, the gap is still substantial in April, notwithstanding a clear reduction since. The gap has nearly completely disappeared in Germany, while it persists, to variable extents, for the other countries. Interestingly, the perception gap is not fundamentally different when national CPI inflation is used instead of HICP inflation. Inflation misperception therefore seems to be a relatively widespread phenomenon and not just an issue characterising the HICP. From the point of view of assessing the accuracy and the credibility of the HICP specifically, this is a rather favorable finding. Results are also robust to an extension of the sample period, which was possible for national CPI inflation (starting in 9 instead of 99). The theory according to which inflation perceptions are based on frequently purchased products is only marginally verified, in the sense that the gap for such an alternative inflation measure is marginally smaller than for overall inflation. As a matter of fact, it is smaller for some countries and larger for others. Moreover, for the past (period before the changeover) perceptions do not track this alternative inflation measure. This theory therefore does not seem to fit the data. We have provided some evidence suggesting that real house price developments could have contributed, to some extent, to the persistent discrepancy between perceived and actual HICP inflation. Further research is however needed in order to assess the precise