Perceived Inflation and the Euro: Why High? Why Persistent?

Transcription

1 Perceived Inflation and the Euro: Why High? Why Persistent? Helmut Stix Abstract This paper employs an Austrian micro-dataset to analyze why inflation perceptions became disconnected from official inflation measures in the course of the euro cash changeover. We find evidence that persons who are confronted with prices of frequently purchased goods, who expected price increases and who mentally convert euro prices into old currency prices when making price comparisons have a significantly higher perception of inflation. Furthermore, our results indicate that the latter two factors have a persistent impact. This contributes in explaining why price perceptions have not normalized for several years in some countries. The results suggest that policy measures in countries which are going to introduce the euro should address these issues in order to prevent a similar development as experienced in many euro area countries. JEL-Code: E31, E58 Keywords: Perceived Inflation, Euro Cash Changeover, Survey Results We would like to thank Erich Kirchler and Manfred Fluch for helpful comments. The views expressed in this paper are strictly those of the author and do not necessarily reflect the views of the Oesterreichische Nationalbank. Oesterreichische Nationalbank, Economic Studies Division, P.O. Box 61, A-1011 Vienna. Helmut Stix: helmut.stix@oenb.at, Phone.: [+43]

2 1 Introduction It is now a well established fact that the euro cash changeover has been accompanied by perceptions of strong price increases. At the same time, official inflation indices have shown only moderate price developments. The wedge between what was measured on the one hand and what was perceived by consumers on the other hand, first, can be observed in most member countries of the euro area and second, has been sizeable. Third and somewhat surprisingly, the gap has turned out to be very persistent. How can this development be explained? The literature has provided various answers. Some focus on the difference between how consumers perceive inflation and the way official indices reflect price changes (Brachinger, 2006). Other explanations are directly related to the euro cash changeover: The role of (difficult) conversion rates is highlighted by Dziuda and Mastrobuoni (2005) and Ehrmann (2006). Kooreman, Faber and Hofmans (2004) show that charity donations increased after the euro cash changeover and attribute this to the presence of money illusion. Another reason could be that a sizeable fraction of the population still compares euro prices with legacy currencies prices that, in the meantime, are four years old (e.g. Mastrobuoni, 2004). In contrast, Traut-Mattausch, Schulz-Hardt, Greitemayer and Frey (2004) attribute high inflation perceptions to the widespread existence of expectations of price increases prior to the cash changeover. Del Giovane and Sabbatini (2005b) additionally analyze the importance of media coverage and the effect of a coincident deterioration of the income situation of some socio-demographic groups. These approaches provide interesting insights. Nevertheless, the underlying hypotheses have either not been empirically falsified or have only been indirectly confirmed (in experiments, in a descriptive way or through cross-country comparisons). And if the hypotheses have been tested, most papers have only tested them in isolation omitting other potentially relevant explanations. The present paper extends the literature by providing direct evidence from individual data in particular a survey conducted in Austria in summer 2004 about perceived inflation (commissioned by the Oesterreichische Nationalbank). 1

3 This survey allows for a judgment about what factors drive price perceptions on the level of an individual and how people think about price increases. Furthermore, by allowing to test for several hypotheses simultaneously we can discriminate between them. Why should we care? We think that understanding the reasons behind the increase in perceived inflation is important for several reasons some of them going well beyond the specific case of the euro changeover in 2002: First, if a currency conversion is associated with the nimbus of prices increases, then this might undermine the credibility of official price measures and/or aversely affect public support for the new currency. Empirical evidence of both effects is reflected in survey responses, which for example indicate that the share of those saying that they view the new currency negatively is much higher for those that have perceived strong price increases (although, admittedly, the direction of causality is ambiguous). Second, increased price perceptions might have real effects. For example, this might be the case if consumers overestimation of inflation results in an underestimation of their purchasing power, causing suboptimal consumption decisions (ECB, 2002). Experimental results of Hofmann, Kamleitner, Kirchler and Schulz-Hardt (2006) provide supportive evidence for this conjecture. Furthermore, Janger, Kwapil and Pointner (2005), who conducted a survey among Austrian individuals about (weak) consumption spending in 2004, find that consumers name higher prices as the most important reason why they spent less. Third, price perceptions might also have an impact on the formation of inflation expectations. Evidence that inflation expectations grew with inflation perceptions is provided in Fluch and Stix (2005). And finally, a profound knowledge about the factors affecting inflation perceptions is important in light of the forthcoming introduction of euro cash in some of the new member states. The paper is structured as follows. The literature which explains this development is reviewed in Section 2. Our hypotheses are discussed in Section 3. In Section 4 we present the empirical model and in Section 5 our results. Section 6 concludes. 2

4 2 Why Have People Perceived High Inflation Rates? Results from the Literature In this section we will briefly and selectively summarize the main arguments that have been given in the literature to explain the wedge between perceived and actual inflation. We will first focus on explanations about how consumers realize price changes and then will discuss some explanations that are specific to the euro cash changeover. 1 It has been stipulated by various authors that the level of perceived inflation could be positively correlated with price increases of frequently purchased goods (e.g. ECB 2002). A formulation of this idea is provided in Brachinger (2006), utilizing elements from Prospect Theory. In particular, the theory maintains that price changes of more frequently purchased goods are weighted more by consumers than price changes of less frequently purchased goods. Furthermore, it is assumed that consumers weight price increases more strongly than price decreases. 2 As prices of more frequently purchased goods have increased more than prices of less frequently purchased goods in many countries of the euro area (cf. Del Giovane and Sabbatini, 2005a; ECB, 2003), the theory predicts an increase in perceived inflation. 3 In an empirical application of his theory, Brachinger (2006) reports that perceived inflation was about four times higher than HICP-inflation around Based on specific parameter assumptions, this index also indicates that, in contrast to evidence derived from surveys, perceived inflation in Germany has not faded and still deviates substantially from measured inflation. 4 In contrast, Aucremanne and Collin (2006) find, in an aggregate time 1 We will neglect the role of psychological prices (e.g. el Sehity, Hoelzl and Kirchler, 2005; Mostacci and Sabbatini, 2005; Fluch and Stix, 2005) and other explanations that are mainly country-specific. 2 More specifically, each consumer has a value function. For each product, the consumer evaluates his value function where the evaluation is asymmetric with respect to price increases and decreases. The loss aversion parameter is assumed to lie in a range from 1.5 to 2.5 price increases are perceived 150 to 250% as strong as price decreases. 3 For Austria, Haschka (2004) finds that a price index consisting of a basket of (typically) daily purchased goods, increased on average by 3.3% from 2001 to 2004, while one consisting of a typical weekly purchase increased by 2.5%. In comparison, the HICP grew by only 2%. 4 For a critique on Brachinger (2006), see Hoffmann, Leifer and Lorenz (2005). 3

5 series model, no evidence that the price development of frequently purchased goods and services is important in explaining the perception gap. 5 It is well known that prior to the changeover the fear of price increases was very widespread. According to a Flash Eurobarometer survey from November 2001, 70% of the euro area population were afraid of price cheating; for Austria this figure was more modest (52%). 6 The impact of these expectations has been demonstrated by Traut-Mattausch et al. (2004) who have conducted several experiments where German probands received menus in Deutsche mark and afterwards menus in euro. They found that price increases were overestimated even when prices were converted correctly. When euro menu prices were too low, price increases were estimated to be zero and when euro prices were too high, the price increases were estimated to be even higher. Thus, evidence of illusionary price increases due to the euro introduction is found in spite of the fact that clear disconfirming evidence was available (Traut-Mattausch et al., 2004, p. 756). The authors assign this effect to a selective error correction mechanism. 7 Hofmann et al. (2006) have repeated this experiment in 2004 for Austria and, even after two years of experience with the euro, obtained similar results. Furthermore the experimental setting has been extended to wages, which were to a large extent perceived correctly. Therefore, the authors conclude that a combination of higher perceived inflation and unchanged wages can result in a subjective loss of purchasing power, probably affecting consumption decisions. When conducting price comparisons, the reference values which are used by individuals are of great relevance. In this context it is very important that legacy currency prices have been widely used as reference prices for judging price developments. For example, 5 It should be noted that different authors collect very different goods and services under the heading frequently purchased. This complicates a comparison of results. Probably, the calculation of a true price index for daily or weekly goods can be regarded as best practice in this context (see fn 3). 6 Flash Eurobarometer 11/2001, Question 8: You re afraid of abuses and cheating on prices. 7 Accordingly, errors that are consistent with expectations are less likely to be realized than errors that run counter to expectations. For example, if prices are mistakingly overestimated then this error will less likely be detected by a person which had expected price increases than by a person which had no such expectations. 4

6 Fluch and Stix (2005) show that as late as in the summer of 2004, almost 40% of all Austrians still always or frequently converted euro prices into Austrian schilling. Since the Austrian schilling prices are now several years old, the level of perceived inflation increases with the temporal distance from the schilling area just due to the normal inflation process. Brachinger (2006) also accounts for the widespread use of legacy currency prices in a variant of his index of perceived inflation by assuming that people estimate price changes relative to a moving reference period which both contains Deutsche mark and euro prices. Another aspect of the conversion has been highlighted by Mastrobuoni (2004) and Dziuda and Mastrobuoni (2005). In particular, these models build upon the idea that a new currency decreases transparency of prices, i.e. consumers are not used to the new currency and recognize prices, due to difficult conversion rates, only with some error. For retailers, this generates an incentive to increase prices. Since the costs of erroneous conversions that arise to consumer are small for goods with a low price, the model predicts that the incentive to increase prices is inversely related to the initial price level. This aspect of the model is tested for EMU and non-emu countries and an inverse relationship between the price level and inflation can be observed in some EMU-countries. 8 Also, it is stipulated that price increases should be correlated with consumers ability to adapt to the new currency. Dziuda and Mastrobuoni (2005) test for this effect indirectly by approximating consumers ability by the (aggregate) share of the population which uses old currency prices when making price comparisons and by the share which feels uncomfortable with the euro. In a cross-country regression both variables seem to be correlated with the size of the inverse relationship between inflation and the price level. 9 8 Another aspect of the model is that the strength of this inverse relationship should be positively related to the market power of retailers, which can also be confirmed empirically. Gaiotti and Lippi (2004) assembled a panel of 2500 restaurants in Italy and also find that local market power was associated with a larger price increase. They also propose a theoretical model for this observation. In contrast, Hobijn, Ravenna and Tambalotti (2004) use a sticky-price model to argue that the increase in restaurant prices can be explained by menu costs. 9 In particular, Dziuda and Mastrobuoni (2005) run both a country-wise and a panel regression to estimate the relationship between the price level of goods and the inflation rate. In the latter case, they find a negative correlation for almost all EMU countries, as predicted by the model. In the former case of country-wise regressions, however, the effect is significant only for a few countries. In particular, it is not significant for Germany and Austria, which is somewhat surprising given the evidence presented in our 5

7 In a similar direction, Ehrmann (2006) argues that the new denomination has increased the information-processing requirements of consumers namely, by more than for sellers which potentially can lead to price increases. Hence, in an empirical model, Ehrmann (2006) tests whether the complexity of the conversion rate has affected inflation in In a cross-country sectoral analysis he finds evidence in favor of his theory, but only for low-priced goods sold in mid-priced stores in some sectors. Interestingly, the complexity of the conversion rate seems to affect inflation and inflation perceptions nonlinearly. In particular, the inflation perception gap seems to have been higher in countries with a medium-complex conversion rate, whereas a lower effect is found in countries with simple and complex conversion rates. Furthermore, the perception gap declined faster in countries with a simple conversion rate. Additionally, other euro cash changeover specific factors have been discussed: Del Giovane and Sabbatini (2005b) analyze the importance of media coverage about price increases and the fact of a coincident deterioration of the income situation of some socio-demographic groups. Aucremanne and Collin (2006) find that differences between national price indices and the euro area Harmonized Price Index (the latter excludes owner occupied housing) can not explain the perception gap. Ehrmann (2006) and Eife (2005) report that legally imposed dual pricing had a dampening effect on prices. To sum up, we think that the discussed hypotheses are very important in explaining the wedge between actual and perceived inflation. Furthermore, their empirical plausibility has been shown in various ways: some explanations are consistent with price movements (in particular some of the propositions from Brachinger 2005) or with cross-country differences in consumers ability to adapt to the euro or in the complexity of conversion rates (Dziuda and Mastrobuoni s and Ehrmann s hypotheses). Other hypotheses have been confirmed in experiments (Traut-Mattausch et al., 2004; Hofmann et al., 2006). In some of the above mentioned studies, aggregate measures derived from survey data are used as proxies of individual behavior for focal variables like consumers ability to paper. 6

8 adapt to the euro. As many of the discussed hypotheses are concerned with individual behavior, we think that it is natural to test the questions at hand by using direct evidence from individual data. This is missing in the literature. In particular, the use of survey data would allow to answer the following questions: Can some of the results also be confirmed on an individual level when all other effects are controlled for? Which explanations are more important? Is it possible to identify the reasons why the disparity was so persistent? Of course, the use of survey data also has disadvantages in comparisons to, e.g. crosscountry comparisons. For example, little can be said about the temporal evolution of the perception gap. 3 Data and Hypotheses To shed some light on these questions we will utilize data from a representative survey conducted in July and August 2004 among the Austrian population persons above the age of 15 were interviewed by telephone on questions ranging from an assessment of whether prices have changed and the specific reasons therefor, to difficulties with euro conversion and the attitude and assessment of the euro. This provides the opportunity to test for the relevance of some of the hypotheses raised in the literature among them hypotheses which appear most prominently in the policy debate. 10 However, it should be noted that the survey was not specifically designed for the purpose of this paper which somewhat limits its usability (the main focus of the survey was on psychological factors in relation to the euro). In some cases, we will therefore need to transform some survey answers in order to make them usable for our analysis. The three hypothesis which will be tested are: First, we deal with the way how individuals handle the presence of a new unit of account and how they form their individual price perception. The available information from the survey allows to approach this from several directions. The first deals with the 10 See the Appendix for a detailed description of the survey and the questions. Summary statistics of the variables used are summarized in Table 1. 7

9 conversion into the old currency. If individuals still convert euro into Austrian schilling, then they will perceive higher prices because they compare actual prices with prices that are frozen as of We can address this issue because the survey contains questions about whether respondents convert into schilling and how regularly they do so. Specifically, the corresponding answers are grouped into those that always ( CONVERT A ), frequently ( CONVERT F ) or occasionally ( CONVERT O ) convert and those that rarely or never convert into schilling. That this effect can be of relevance is reflected by the fact that two and a half years after the cash changeover, still 13% of the population converted always and 27% converted frequently. Further 34% did so occasionally. 11 Another set of variables deals with the way how prices are converted. The correct conversion rate for the Austrian schilling is 1 (EUR) = (ATS). Obviously, this conversion rate is difficult to handle in day-to-day situations and hence the typical approximation used is 1:14. However, this approximation can be very misleading as it implies an inflation rate of 1.7% which, for example, is higher than the annual HICP-inflation rate in In fact, 61% of the Austrian population fully agree to the statement when converting from euro to Austrian schilling, I round tolerantly and convert with 1: In the regressions we make use of this information by including a dummy variable for this answer ( CONVERT 1:14 ). An alternative way to cope with how people convert is to directly measure respondents conversion ability. In particular, survey participants were asked to spontaneously convert the amount of 1.80 euro into schilling. For those that overor underestimate the correct amount by 10% we define the dummy variables CONVERT +10% and CONVERT -10%. 13 Our second hypothesis is related to the role of expectations, in particular to the finding 11 It is clear that the use of dummy variables is a gross simplification of the often complex strategies to learn new prices. 12 The use of the word tolerantly represents a value judgment, possibly biasing the results towards higher price perceptions. 13 The dummy variables CONVERT 1:14, CONVERT +10% and CONVERT -10% take a value of one only if survey respondents actually converted (at least occasionally). Hence, the inclusion of these dummy variables in the regressions allows for testing whether respondents who convert with errors have higher inflation perceptions than respondents who convert without errors. 8

10 that widespread expectations of price increases prior to the euro cash changeover are related to the perception of price increases. The survey contains a question about the attitude towards the euro before the cash changeover. The question was Which attitude towards the euro did you have prior to the euro s introduction?. The answers are grouped into those with a positive ( ATT POS ), a neutral and a negative attitude ( ATT NEUTRAL, ATT NEG ). About 37% of the population had a positive, 30% a neutral and 33% a negative attitude. As this question is not directly related to the expectation of price increases we alternatively use the results from another question which directly deals with expected price increases: 14 Do you agree/disagree with the following statement: It was clear to me before the introduction of the euro that prices will increase. This variable is labeled EXP. INCREASES. 15 Here, the survey shows that a majority (55%) agreed, while 37% disagreed. The rest had no clear opinion. Regardless of which of these question is selected, it is evident that the use of recallquestions is not optimal as the ability of respondents to remember their attitude of the year 2002 might be affected by respondents perception of prices, i.e. the variable might be endogenous. 16 In principle, one could circumvent this by applying some form of instrumental variable approach, however, this is difficult to handle in the context of an ordered response model. In lack of an alternative, we will follow a pragmatic route by testing whether the inclusion of these recall variables affects the other estimated coefficients and standard errors. Furthermore, there are two facts which suggest that endogeneity might not pose a serious problem. First, we can cross-tabulate the answers about the current attitude (at the time of the survey) with the answers on the recalled attitude (referring to the time prior to the cash changeover). This exercise yields that 59% of all respondents had a different view about the current situation than they had before the euro cash changeover. This suggests that the majority of persons differentiated in their answers about the current 14 The question on the attitude towards the euro captures a general assessment which, however, is very likely to be positively correlated with expected price increases. 15 Respondents gave a grade from 1 ( fully agree ) to 5 ( do not agree at all ). 16 In particular, if a person has perceived price increases, then this might affect the person s answer about his attitude before the cash changeover. 9

11 and the recalled attitude. Moreover, for the recall-question on expected price increases, we are able to check our results with an external data source, in particular a survey by the European Commission which was conducted in November Although the questions are not identical, the answers are qualitatively very similar indicating that, on aggregate, people remember well (or reveal correctly) their past attitude. 17 The third hypothesis: Brachinger (2006) has hypothesized that perceived inflation should be higher if prices of frequently bought goods or services increase more than prices of less frequently bought goods and services. Since we do not have information on the frequency of purchases of individual persons, we will not be able to test for this effect directly. However, the survey allows us to follow an indirect approach. In particular, we can utilize information on whether a person runs a household ( HOUSEHOLD ), i.e whether a person is responsible for daily purchases. As these prices grew more strongly than the prices of less frequently purchased goods, we expect that persons who run the household perceive a higher inflation rate than persons who are not responsible for running the household. Clearly, our crude dummy variable does not allow for a rigorous test of this hypothesis. Hence, we will treat results as indicative: if a significant effect is found then this provides some evidence in the direction of the hypothesis. 4 Model To test for the influence of the above mentioned variables we regress several individual characteristics on two separate variables which are assumed to proxy inflation perceptions. Against the background that perceived inflation is intrinsically difficult to measure and that better measures are not easily available from the survey we think that these two dependent variables serve our purpose of proxying perceived inflation sufficiently well According to the EC survey from November 2001, 52% of all Austrians feared abuses and cheating on prices in connection with the euro cash introduction. According to the recall question we use, 55% of all Austrians said that they expected price increases. 18 See the Appendix for a discussion on measurement issues, the problems with various methods and a definition of variables. 10

12 The first dependent variable is a binary variable which approximates whether individuals perceive a higher inflation rate than officially measured. This variable is constructed from answers on a direct quantitative estimate of the level of inflation survey participants were asked for. We find that individuals estimate an inflation rate of 5% on average, which is more than twice as high as the inflation rate in August 2004 (2.3%). Due to the sequencing and the exact wording of the question we will not use answers on this question directly in our analysis. The reason is that respondents were first asked whether the have heard about the term inflation rate. If so, they were subsequently asked for the level of inflation and if respondents didn t know the answer they were asked for an estimate. Unfortunately, the data set does not allow to extract ex-post whether individuals actually gave an estimate or whether they knew the level of inflation, making it impossible to determine the extent to which answers reflect perceived rather than measured inflation. 19 However, this problem can be circumvented by converting the quantitative estimate of inflation into a binary variable. In doing so, we utilize information on respondents estimates of inflation together with answers on a question about the credibility of the inflation rate. In particular, respondents were asked whether the inflation rate can truthfully represent the price development. 20 This allows to construct a variable, labeled PERCEP- TION, which indicates whether an individual perceives an inflation rate that is higher than the measured rate. The variable is constructed as follows: If a respondent answered that the level of inflation is higher than the measured rate of inflation then PERCEP- TION takes a value of one. Those respondents which provided an answer which is close to the CPI-inflation rate either did (at least approximately) know its level or perceived a rate close to the measured rate. To distinguish these cases, we use information of the assessed credibility. In particular, if these respondents answered that the inflation rate 19 In particular, it could be the case that individuals perceived a higher level of inflation than indicated by the official inflation measure but nevertheless replied the official measure because they knew its level. 20 The exact question is: Do you believe that the actual price development can be truthfully represented with (the rate of inflation)? I mean, how credible (is the rate of inflation) in your view? Please indicate. Respondents gave a grade from 1 (very credible) to 5 (not at all credible). Further details on the construction of the dependent variables can be found in the Appendix. 11

13 is not credible, then their inflation perception must be different from their answer and PERCEPTION also takes a value of one. In turn, those who say that it is credible have no perception gap ( PERCEPTION = 0). Finally, all persons who estimate the inflation rate below the measured level and think that the rate is credible also do not have an (upward biased) perception gap. The only ambiguous group consists of those respondents who estimate the inflation rate below the measured rate but think that the inflation rate is incredible. From the latter information, we know that these person s perceived rates are higher than what they answer but we do not know whether their perceived rate is higher than measured inflation. In the baseline case, we assume that those giving an estimate below the measured rate and think that it is incredible are also assumed to have a perception gap. 21 In practice, it seems sensible to provide a range around the estimated inflation rate which defines what can be viewed as a relatively precise estimate. We have chosen a range of plus and minus 0.5% around the true rate (from 1.8% to 2.8%). 22 It is important to note that the sequencing of questions implies that this dependent variable is defined for only those that have heard about the term inflation rate which applies to 97% of respondents. Among these, 78% provide a quantitative estimate. Hence the sample defined by this dependent variable comprises people with an opinion about the inflation rate. Furthermore, an interesting control group can be constructed if the sample is restricted to only those whose estimate lies between 1.8% and 2.8%. This sample selects only people which are obviously rather well informed about the inflation rate. In the empirical section, we can then test whether psychological effects can be found also for this sample. The survey also asks about the development of prices in a qualitative manner. In particular, the second dependent variable is derived from a question about whether the introduction of euro cash has induced price increases ( PRICE INCREASES THROUGH 21 To check whether this assumption affects the results, we will also report results for alternative definitions of PERCEPTION for this group. 22 The estimates for other ranges, e.g. from 1.5% to 2.5%, yield similar results qualitatively. 12

14 EURO INTRODUCTION ). 23 In total, 56% of all Austrians answered that many products had become more expensive with the introduction of the euro, 34% said that some products had become more expensive. Since only a very small fraction of respondents answered that prices got cheaper, we merge this category with the answers that prices stayed the same, applying in total to 10% of respondents. We include the regressions with the qualitative indicator on how people see price changes as a dependent variables for three reasons: First, the dependent variables cover different time horizons: PRICE INCREASES THROUGH EURO INTRODUCTION refers to price changes over a period of two and a half years (as seen from August 2004). In contrast, PERCEPTION covers the last year. Second, the samples covered by the two dependent variables is different. Whereas PERCEPTION contains a limited sample, answers for the second dependent variable are given by all respondents, irrespective of whether respondents knew or did not know the term inflation rate. And third, both dependent variables are based on a completely different set of questions. Hence, this provides a robustness test of our results which seems very important given the difficulties in measuring inflation perceptions. As the variable PRICE INCREASES THROUGH EURO INTRODUCTION asks for an assessment about how many products got more expensive (and not about an categorical assessment of the level of inflation), the question arises whether this variable is able to provide an (approximate) measure for perceived inflation. As the latter is unobserved we cannot provide a direct test. However, we can conduct several cross-checks. First, we think that the exact wording of the question is such that it entails an ordering. If someone felt that prices increased then she could choose between many and some prices. We believe that in the vast majority of cases people associate the statement that many goods got more expensive with a higher rate of inflation than with the statement some goods got more expensive. Second, we can analyze whether answers on this categorical question 23 The exact wording is: Do you believe that the introduction of the euro was accompanied by price changes. Answers range from yes, many goods got more expensive, yes, some goods got more expensive, no, no change, to yes, some goods got cheaper and yes, many goods got cheaper. 13

15 are correlated with survey participants quantitative estimates of the rate of inflation. This exercise shows that the mean estimate of the rate of inflation is 3.1% for those saying that there were no price increases or products got cheaper, 3.6% for those saying that some goods got more expensive and 6.4% for those answering that many goods got more expensive. Thus, these results suggest that the assumption of a correlation between the estimated rate of inflation and categorical answers on how many products got more expensive is not unwarranted. Given the binary nature of the first dependent variables, the estimation model is specified as a probit, in case of the last dependent variable as the following ordered probit model: yi = β 1 SD i + β 2 E i + β 3 C i + ɛ i (1) where yi represents the inflation perception of individual i, SD i is a vector of various sociodemographic variables, E i is a vector of variables capturing the expectations of individual i and C i is a vector of variables controlling whether individual i converts into schilling and if so how this is done. ɛ i is an error term assumed to be normally distributed with mean zero and unit variance. Instead of observing the perceived inflation rate yi directly, we have data on y i, the categorical survey response of individual i. Iftherearem categories, then y is in the j-th category if it is in the range given by α j 1 <y<α j, where the α s are parameters to be estimated Estimation Results Tables 2, 3, and 4 summarize the regression results for the three different dependent variables (the binary PERCEPTION variable both for the full sample and the control group and the categorical perception measure for the variable PRICE INCREASES THROUGH EURO INTRODUCTION ). For each dependent variable we estimate several specifications. 24 The model is estimated by maximum likelihood. 14

16 Studies have shown that the socio-demographic characteristics of a person are important determinants of inflation perceptions (e.g. Bryan and Venkatu, 2001; Lindén, 2006; Jonung, 1981). For this reason, we add dummy variables for six household income classes ( INC ) as well as for the age of respondents ( AGE ). The idea is that price increases might have a different impact for a high income household than for a low income household because of liquidity constraints (a low income household spends most of its income with little variation left for savings; the relative costs of price increases are thus higher for low income households). Alternatively, one could argue that the composition of the consumption basket differs between households. For example, richer households might consume more services than poorer households. If service prices increased more than food prices then this line of reasoning would imply that richer households face a higher perception gap. Also, it could be argued that age is important. For example, retired people with low nominal increases of their pensions are more affected by perceived price increases than younger households who can adjust labor supply. On the other hand, one could argue that old persons, specifically those who are retired, have more time to get informed about price movements than young people. Another set of dummy variables controls for the educational level of a person ( EDU ). We control for education first because this provides additional information on personal income, which is not available from the survey data, and second because the level of education might have implications on perceived inflation directly. For example, less educated people might have less information on price movements than more educated people. Finally, a robust finding of the literature is that females perceive higher inflation rates than man. Therefore, we will also control for the sex of a person. Factors determining price perceptions The first set of results, shown in Table 2, summarizes the marginal effects on the probability that a person has a perception gap. The findings in column 1 of Table 2 imply that age and the likelihood of an inflation gap follows an U-shaped pattern: young persons below the age of 25 (the reference group) have the highest probability of a perception gap. Relative to those aged 25 to 34 years, the probability is about 30 percentage points (Pp.) higher. This result can possibly be 15

17 explained by the fact that young persons estimates of inflation are very high (the mean rate of estimated inflation is 7.3% for those aged below 25 and 3.4% for those aged 25 or above). 25 Therefore, it seems likely that young persons are not informed or probably do not want to be informed about price changes, hence we do not assign much significance to this result. More important are the relative effects among age groups older than 24. Here we find that the likelihood of a perception gap increases with age people aged 55 or over face a probability that is 18 percentage points higher than persons aged 25 to 34. Expectedly, education exerts a very clear influence with the likelihood of a perception gap decreasing with the level of education. In particular, the point estimates imply that someone with an apprenticeship ( EDU APP. ) has a probability which is 10 Pp. lower than someone with only a mandatory educational level (the reference group). A high school or a university degree further decreases the likelihood by 20 and 30 Pp. respectively relative to the reference group. 26 The point estimates from the household income dummies do not allow to identify a clear pattern. First, a Wald test cannot reject the null hypothesis that all income dummy variables are jointly zero. 27 Second, the individual point estimates reveal that the likelihood of a perception gap increases with household income up to 1850 euro and decreases then. However, from an economic point of perspective, the effects are not sizeable. 28 On the one hand, this indicates that household income alone is not much of importance in determining inflation perceptions. On the other hand, viewing the results of the household income variables together with the results of the education variables, which proxy personal income, suggests after all that income plays some role to what extent prices increases are perceived in particular for the highest educated or highest income groups against all other income 25 We have also found very high inflation estimates by young people when analyzing CCB data (unpublished). 26 Bryan and Venkatu (2001) report a similar signed effect of education for US survey data on reported inflation rates. 27 β INC1,100 INC1,500 =...= β INC>2,900 =0. Prob(χ 2 6 > 7.89) = For example: β INC1,100 INC1,500 = β INC2,200 2,900. Prob(χ 2 1 > 4.10) = The difference in the probability that someone from the lowest household income group (relative to someone from a higher income group) faces a perception gap is 8 Pp. at most. 16

18 groups. At the same time, the result also indicate that the perception of price increases is prevalent across a wide range of the population, given that persons without a university degree constitute about 86% of the sample and that people whose household income is not in the highest income group constitute about 75% of the sample. Next, we turn to our hypotheses: First, we test whether the perception of inflation is different for persons who run the household than for persons who do not run a household. As the literature has documented that females have higher inflation perceptions than males, we have interacted the information on whether a person runs a household with the variable controlling for the sex of a person. In the estimated equation three dummy variables are added: for females and males who care for daily purchases ( FEM-HOUSEH., MALE- HOUSEH. ) and for females who do not care for daily purchases ( FEM-N-HOUSEH. ). The point estimates of these variables measure the effect relative to males who do not care for daily purchases. The regression results show that males who run the household have a higher likelihood of a perception gap than males who do not run the household (by 8 Pp., respectively). The point estimates also indicate a sizeable effect of 13 Pp. for females running a household versus females not running a household. This effect however is not statistically significant, seemingly attributable to the fact that the standard error for FEM-N-HOUSEH. is relatively large while the point estimate for FEM-HOUSEH. is rather precisely estimated. This might be explained by the fact that the former group comprises only 53 observations. Next, we turn to the variables which indicate whether people convert and if so how they convert into schilling. First, the results do not only indicate that conversion per se matters but also that the frequency with which a person converts is important. In particular, we find that if prices are always converted then this results in a higher likelihood of perceived inflation than if prices are frequently converted. And if prices are frequently converted then this results in higher perceived inflation than if prices are only occasionally or not converted the probability of the event perception gap is found to be higher by 14 Pp. and 10 Pp. respectively for persons who always or frequently convert. Also, the results suggest that the use of a conversion rate of 1:14 does not significantly increase the likelihood of a 17

19 perception gap (extra to what has been captured already in the dummy variables for the frequency of conversion). Finally, we turn to the results for the variables which measure the attitude towards the euro. It is found that persons who had a negative or neutral attitude before the changeover have a higher inflation perception than persons who had a negative attitude. In contrast, no significant effect is found for persons who expected price increases before the cash changeover. However, it is likely that these two sets of variables are collinear. Therefore, columns 2 to 5 of Table 2 summarize the results of several different specifications. To account for the correlation between the variables measuring the attitudes before the cash changeover ( ATT NEGATIVE, ATT NEUTRAL ) and the variable measuring expected price increases ( EXP. INCREASES ), we alternately omit one of them in column 2 and 3. We find that the omission of one variable renders the other significant, pointing to the presence of multicollinearity. The size of the effect of these group of variables varies between6and8pp. Due to the possible endogeneity of these variables, we omit them altogether in column 4. The fact that the results do not change qualitatively suggests that endogeneity might not pose a problem. Finally, in column 5 we omit the variable measuring the 1:14 conversion, which rests on self assessment of the participants in the survey, and include the variables which measure if respondents over- or underestimated the euro amount of 1.80 by 10%, respectively. The coefficients have the wrong sign and only one coefficient is marginally significant. Hence, we tend to interpret this as a suggestion that the imprecision of conversion does not add to the likelihood of a perception gap beyond what is not already captured by the frequency of conversion ( CONVERT +10% and CONVERT -10% are only defined for those who convert). As mentioned, the dependent variable is constructed by assuming that persons with a quantitative estimate for the rate of inflation below 1.8% and who answered that the inflation rate is incredible in truthfully representing price developments have a perception gap (i.e. that the variable PERCEPTION takes a value of one). In order to test for the robustness of our results we have also repeated the analysis with two alternative definitions 18

20 of the dependent variable. First, instead of caring about how to treat those with an inflation estimate below 1.8%, we exclude all these respondents from the analysis. This reduces the sample size by about 500 observations. Second, we have excluded only those respondents with an estimate below 1.8% who also believe that the inflation rate is incredible. This reduces the sample size by about 350 observation. Reassuringly, we find that the results from both regressions do not change qualitatively. 29 It could be argued that the presence of high inflation perceptions and or the presence of psychological effects are only attributable to a small share of the population which has little knowledge about or interest in the inflation rate. Thus answers from this group could drive the results and mistakingly render the impression that euro factors can be found for the whole population. To account for this argument, another robustness test has been conducted by considering only those respondents with an inflation estimate between 1.8% and 2.8%. As this group consists of persons who are rather well informed about the inflation rate it provides an interesting control group for the above mentioned argument. The results, shown in Table 3, reveal, as in the regression with all observations, a U-shaped pattern concerning the effect of age, a significant effect of education, and a small effect of household income. Concerning the three hypotheses, we find that persons who care for daily purchases do not have a statistically higher likelihood of a perception gap. 30 In contrast, the effect of expected price increases and the attitude towards the euro remain unchanged qualitatively. Furthermore, we again find that mental price comparisons in old currency do affect the results. In this case, those who convert frequently have a significantly higher probability of a perception gap than others. 31 Thus, the results from this control group demonstrate that euro related psychological factors are important even for respondents who are well informed about the price development. 29 The results are not shown but are available from the author. 30 Females who care for the household have a higher coefficient than females not caring for the household. This effect is again not significant statistically, however. 31 The point estimate of CONVERT ALW is positive but not significant. This might be due to the fact that this group is very small and consist of only 67 observations (about 10% of the sample) which is not much given the high number of dependent variables. In contrast, the group CONVERT FRE consist of about 200 observations. 19

21 Factors determining price perceptions caused by the euro cash changeover In a next step, the analysis is repeated with PRICE INCREASES THROUGH EURO IN- TRODUCTION as the dependent variable (Table 4). The analysis is thus focused on a longer time horizon. As the responses of this variable are ordered from no change to many goods got more expensive, a positive sign of the coefficients β indicates that a variable positively affects the probability that individual i perceives more price increases, or as we assume, a higher rate of inflation. 32 As before, the findings in column 1 of Table 4 reveal a significant influence of education, with the level of perceived inflation decreasing with the level of education. In particular, the point estimates imply that the probability that a person answers that many goods got more expensive is higher by 13 Pp. and 9 Pp. if this person does not have a university degree or a high school leaving certificate, respectively. Also, the results for the household income dummies are rather similar than in the previous regressions with some effects between income groups, in particular for the highest against some middle income groups, but no sign of a sizeable impact. 33 Concerning the effect of age we find little variation between age groups indicating first that for a longer horizon perspective on perceived inflation age does not seem to matter. 34 Second, in contrast to previous results we do not find that persons aged 24 or lower have a significantly lower inflation perception which is probably caused by the qualitative nature of the dependent variable which does wash out implausible answers by the young. Next, we turn to our hypotheses: First, females who run the household have a significantly higher likelihood of inflation perception than females who do not run the household (by 10 Pp.). 35 In contrast, males have no different inflation perception irrespective of whether they care for the household or not. This again provides some evidence that per- 32 The coefficients in Table 4 show the marginal effect on the index function. This marginal effects have the same sign as the marginal effects on the probability of the outcome many goods got more expensive. 33 The maximum effect between household income groups is again around 8 Pp. 34 The only significant difference can be found between those aged 45 to 54 and those with an age above Hypothesis test for the equality of β FEM HOUSEH. = β FEM N HOUSEH. yield p-values between 0.04 and 0.06 depending on the specified model (columns 1 to 5). 20