Does the Phillips curve hold for consumer survey data? 1

Transcription

1 EUROPEAN COMMISSION DIRECTORATE GENERAL ECONOMIC AND FINANCIAL AFFAIRS Policy strategy and co-ordination Economic situation, forecasts, business and consumer surveys Does the Phillips curve hold for consumer survey data? 1 Introduction Expectations play an integral role in economic theory. Consumers usually form their expectations about macroeconomic variables when they make decisions with long-term importance, for example, when planning a major purchase (buying a house). Their expectations about future prices, incomes, taxes, interest rates or other variables impact their choices, which eventually influence the overall economic activity via the price mechanism. This is the reason why economists study consumers' expectations. Consumer surveys are utilized to collect data about consumers' expectations for monitoring, forecasting and research purposes. This data provide various opportunities for economic research. It is sensible to study patterns between different consumers' expectations or to analyse the relationships between relevant soft and hard data. One of the research opportunities is also to check whether consumers' expectations reflect economic concepts. Dräger, Lamla and Pfajfar (2013) evaluate whether US consumers form macroeconomic expectations consistent with three different economic concepts: the Phillips curve, the Taylor Rule and the Income Fisher Equation. Based on their analysis of the microdata of the Michigan Survey of Consumers, they conclude that "consistency with economic concepts on average moves consumers' inflation forecasts closer to professionals' estimates." Their research and findings provide the inspiration for this paper, which attempts to reproduce a similar inquiry on European consumers' expectations, albeit focusing on only one of the above-mentioned economic concepts. The objective of this paper is to analyse European consumers' expectations about the future developments of inflation and unemployment and check whether they are consistent with the Phillips curve a concept, which describes the relationship between the inflation rate and the rate of unemployment as an inverse relationship. The Phillips curve (PC) was proposed by A.W. Phillips (1958) in his research paper based on the empirical observation and analysis of British data of wage inflation and the level of unemployment in the period from 1861 to The PC has been a controversial macroeconomic concept from its origin despite its initial success. Phelps (1967) and Friedman (1968) argued that the PC is vertical in the long-run because the price changes do not impact the level of unemployment, which in the long run equals the natural rate of unemployment. Monetarists started to distinguish between the long-run and short-run PCs. The short-run curve, also called the expectations-augmented PC, began to take account of workers' adaptive expectations of price 1 Prepared by Marek Doval and Roberta Friz (DG ECFIN). The views expressed in this study are those of the authors and do not necessarily reflect those of the European Commission.

2 inflation. The stagflation of the 1970s confirmed the non-existence of a stable negative relationship between the inflation rate and the rate of unemployment in the long-run. According to Gordon (2011) research on the PC developed in two separate strands since The first is represented by a classical model, which considers demand and supply factors as well as inflation expectations, whereas the Keynesian PC, which takes into account only forward looking inflation expectations either in its traditional or hybrid version, presents the other line of research. The present note focusses on the consistency of consumers' inflation and unemployment expectations in the European Commission's Harmonised Consumer Survey with the PC. Survey questions 6 and 7 are relevant for this research. 2 Question 6 (Q6) inquires consumers' expectations about future consumer prices developments and question 7 (Q7) refers to consumers' expectations about the future change in the number of unemployed people. 3 To evaluate the theory-consistency of consumers' expectations in the EU, the euro area and other selected major Member States, correlation coefficients are calculated between the time series of the balance statistics of Q6 and the Q7 using both the full sample and shorter periods, which are determined by business cycle phases. Consistency of relevant hard data (i.e. statistical data on inflation and unemployment) with the PC is also examined. In addition, the analysed data are presented in the form of PCs, the relationship between the variables is charted in figures. Last but not least the theory-consistency of consumers is evaluated by taking into account their gender, age, education and income. The analysis confirms the presence of a negative relationship between price and unemployment expectations of EU and euro-area consumers when the whole analysed period is considered. However, consumers in the EU are not strictly theory-consistent (correlation coefficient rather weak), while in the euro area correlation is somewhat stronger. Results for Member States are mixed. Only German and Dutch consumers have theory-consistent expectations with negative correlation coefficients, while the correlation coefficients of the rest of the observed consumers are closer to zero. When shorter periods of expansions and recessions are considered, there are no systemic changes observed in the theory-consistency of consumers related to the changes of the business cycle. This contrasts with the findings by Dräger et al (2013) for the US. The findings for the soft data are broadly in line with those for hard data. However, correlation coefficients obtained for the soft data indicate somewhat better theory-consistency than for the hard data. The conclusions of the consumers' categories analysis can be summarized into one sentence: a young rich man is more theory-consistent than an elderly poor woman. The paper is organised as follows: Section 1 introduces the employed data. The aim of section 2 is to illustrate how the soft data relate to the respective hard data. Then in section 3 the theoryconsistency is examined on the hard data. The results provide a reference point for the evaluation of theory-consistency in the soft data, which is carried out in section 4. Section 5 presents and describes the PCs of the used data, based on both hard and soft data. Section 6, looks at different categories of consumers and evaluates whether there are any systemic changes in theory-consistency of consumers' expectations between these categories. The last section, section 7, concludes. 2 Also the question 4 of the consumer survey, which asks consumers on their expectations on development of general economic situation, can be used to examine consumers' theory-consistency with the PC. Annex 1 contains a brief analysis using the questions 4 and 6. 3 Exact wording of the Q6 and the Q7 is provided in the Data description section. 2

3 1 Data description The aggregate soft data of the European Commission's Harmonised Consumer Survey are employed for the analysis. This monthly survey is conducted in all 28 European Union (EU) Member States (MS) by national partner institutes as part of the Joint Harmonised EU Programme of Business and Consumer Surveys. The nominal sample size for the EU is around respondents and respondents for the euro area (EA). The effective sample size is around respondents for the EU and respondents for the euro area. The survey is designed to be representative at national level. The respondents of the consumer survey are categorized according to six criteria: income, occupation, working regime, education, age and gender. The Harmonised Consumer Survey questionnaire contains altogether seventeen questions, only two of them are quantitative, the rest is qualitative. The qualitative questions 6 and 7 are relevant for the purpose of this paper. Q6 asks about consumers' expectations about future consumer prices developments and Q7 refers to consumers' expectations about future changes in the number of unemployed people. Q6: By comparison with the past 12 months, how do you expect that consumer prices will develop in the next 12 months? They will ++ increase more rapidly + increase at the same rate = increase at a slower rate - stay about the same -- fall N don't know. Q7: How do you expect the number of people unemployed in this country to change over the next 12 months? The number will ++ increase sharply + increase slightly = remain the same - fall slightly -- fall sharply N don't know. Respondents' answers are presented as balance statistics. Aggregate balances are measured as percentage points of total answers and are calculated for each question as the difference between the percentages of positive and negative answers using the following formula: B = (PP + ½P) (½M + MM) PP, P, E, M, MM denote the percentages of respondents who have chosen one option (++), (+), (=), (-) or (--). N is the percentage of respondents who have selected the option "don't know". The middle option E and "don't know" option N are attributed zero weights, whereas the moderate options P and M are attributed half the weight of the extreme options PP and MM. The sum of PP+P+E+M+MM+N equals 100 and the range of the balance statistic B is from -100 to Aggregate survey results for the EU and the euro area are based on weighted averages of the individual country results. Balances per question at aggregate level (country, EU, euro area) are seasonally adjusted. For the analysis, a subset of monthly consumer survey data is utilized. The subset consists of aggregate survey results for Q6 and Q7 and covers the period from January 1999 to May

4 In addition to the total aggregate balances, the survey data are analysed at EU and euro-area level along four respondents' categories (sex, age, education and income). None of the category data are seasonally adjusted. Survey respondents are categorized into four age, three education and four income categories: Age Education Income AG1: years ED1: primary RE1: 1 st Quartile AG2: years ED2: secondary RE2: 2 nd Quartile AG3: years ED3: further RE3: 3 rd Quartile AG4: 65+ years RE4: 4 th Quartile The hard data come from Eurostat. The all-items HICP monthly data seasonally adjusted index (2005 = 100) is used to calculate annual inflation rates in the EU, the euro area and selected MS. 4 For the unemployment rate in the EU, the euro area and selected MS, the total seasonally adjusted monthly average percentage unemployment rate data are utilized. The employed hard data cover the period from January 1999 to May The unemployment rate data for the EU are available only from January To allow a relevant comparison of soft and hard data, the rate of unemployment is transformed into the annual change in the rate of unemployment. The chronology of recessions and expansions established by the Centre for Economic Policy Research (CEPR) Euro Area Business Cycle Dating Committee is applied to derive shorter time intervals from the available full sample of soft and hard data. The CEPR chronology is adjusted and expanded by the introduction of an additional recession period. This recession is defined by the Economic Sentiment Indicator data and lasts from July 2000 to April The following are the periods of recessions and expansions, which are applied for the analysis in this paper: A. Recessions a. July 2000 April 2003 b. January 2008 June 2009 c. June 2011 June B. Expansions a. May 2003 December 2007 b. July 2009 May The selected MS: Germany, France, Italy, Spain, Poland, the United Kingdom, and the Netherlands. 5 The committee members on the last meeting, which took place in June 2014, decided not to call an end of the current recession amid observed weak economic growth in the EA. For the purpose of this paper, June 2014 is considered as the end of the recession. More information about the functioning of the Committee can be found on the CEPR's website ( 4

5 2 Comparison of soft and hard data How well do the time series of consumers' expectations resemble the time series of the corresponding hard data? In this section, the survey data are presented in the graphs alongside their hard data counterparts. From the figures below it is apparent that the soft data are highly correlated with the hard data. Figure 1 contrasts the time series of EU consumers' expectations about future consumer prices (EU Q6) with the time series of the EU inflation rate as measured by HICP. The correlation coefficient of these two series is 0.74, which demonstrates a strong positive correlation. Figure 1: EU Q6 and the EU inflation rate (January 1999 to May 2015) 5.0% EU inflation rate 4.0% EU Q6 (rhs) 3.0% 2.0% 1.0% 0.0% -1.0% r = Figure 2 illustrates the relationship of EU consumers' expectations about the change in unemployment (EU Q7) with the change in the EU rate of unemployment. The correlation coefficient of these two time series is 0.88, which demonstrates an even stronger positive correlation. Figure 2: EU Q7 and the change in the EU unemployment rate (January 2001 to May 2015) Change in the EU unemployment rate EU Q7 (rhs) r =

6 3 Theory-consistency of the hard data with the Phillips curve Consistency of the hard data with the PC is assessed first, before the evaluation of soft data consistency with the PC. This provides more insight into the analysis of consumers' expectations theory-consistency. The firm link between the respective hard and soft data promises comparable theory-consistency of the hard and soft data. For the purposes of this paper hard data and consumers' expectations are considered as consistent with the PC if a correlation is negative and strong. For simplicity, we describe as "strong" a correlation coefficient from an interval of ǀ0.5ǀ to ǀ1.0ǀ. If a correlation coefficient belongs to the interval -0.1 to 0.1, it is considered as zero or no correlation. Anything between strong and zero correlation is considered as weak correlation. Figure 3 illustrates the relationship between the EU inflation rate and the change in the EU rate of unemployment. Visual inspection of the time series suggests that there is no strong inverse relationship between the variables when the whole sample is considered. This is confirmed by a weak correlation coefficient (-0.14) over the period January 2001 to March In spite of this, there are shorter periods where a strong inverse relationship between the variables is apparent. Figure 3: The EU inflation rate and the change in the EU rate of unemployment (January 2001 to May 2015) 5.0% 4.0% 3.0% EU Inflation rate EU Change in unemployment rate (rhs) r = % % % % -1.5 Changes in the rate of unemployment in the EU were relatively small over the period from the beginning of January 2002 until the end of 2005 when unemployment started to decrease until it reached its minimum in the beginning of The development of the inflation rate in the EU was broadly stable around 2% over the period from mid-2002 to mid-2007, just before the Great Recession. Inflation soared between August 2007 and August Then it plunged and reached its minimum around mid While inflation was rising between August 2007 and August 2008, unemployment was still decreasing before it started to grow dramatically from September In the second half of 2011, when the EU returned to recession, inflation began to decline gradually and unemployment continued to grow until mid-2013 when it reached its maximum. Since then both 6

7 unemployment and inflation were on a downward trend, which is not in line with the PC. 6 However, the latest inflation data indicate a bouncing back of the EU inflation rate. Table 1 presents the correlation coefficients for the inflation rate and the change in the rate of unemployment in the EU, euro area and seven selected MS, for the full sample as well as for the shorter periods, which are defined by the business cycle. When the full sample is applied none of the correlation coefficients calculated for the countries, the EU or the euro area confirm consistency of the analysed hard data with the PC. The EU, the euro area, Germany, France and Spain show weak negative correlation, whereas Italian and Dutch correlation coefficients are close to zero. Moreover, Polish and British correlation coefficients are positive and close to the threshold of Considering the periods of recessions (orange) and expansions (blue), there is no systematic pattern in the values of the correlation coefficients related to development of the business cycle. However, in the period of the Great Recession (January 2008 to June 2009) and in the following expansion (July 2009 to May 2011) inflation and the change in unemployment show very strong, almost perfect negative correlation in the EU and the euro area as well as in many of the selected MS. Among the selected MS, Germany is the only country which records negative correlation coefficients in each of the observed periods. Table 1: Correlation coefficients for the inflation rate and the change in the rate of unemployment Country 1999: : : : : :12 Period 2008: : : : : :06 EU EA Germany France Italy Spain Poland United Kingdom Netherlands Note: The full sample for the EU is shorter than for the euro area and the selected MS. The full sample for the EU in Table 1 and Figure 3 covers the period from January 2001 to May Ed Dolan (2014) noticed a similar development in the US data (from mid-2011 to early 2014) in the aftermath of the Great Recession and argues that the period of the US economy recovery from the Great Recession is not consistent with the shifting PC model (monetarists' short-run PC), which contrasts with the previous US recoveries. He concludes: "If the Phillips curve is not dead, we must at least declare it in a persistent vegetative state from which it will not awaken without some radical change in circumstances." (Dolan, 2014) His conclusion is based on the assumption of the shifting PC model, which says that a simultaneous decline in inflation and unemployment can only take place if inflation expectations are falling. 7

8 4 Theory-consistency of consumers' expectations with the Phillips curve This section evaluates theory-consistency of consumers' inflation and unemployment expectations with the PC. 7 Figure 4 depicts the two EU consumers' expectations time series, which cover the period from January 1999 to May Figure 4: Consumers' expectations in the EU - Q6 and Q7 (January 1999 to May 2015) EU Q6 EU Q7 r = Figure 4 reveals substantial information about the relationship of consumers' expectations on price and unemployment changes. There are periods when the series move inversely, but also some periods when they co-move. Co-movement of the series, which contradicts the PC, can be seen since the beginning of 2001 until mid-2002, in the years before the Great Recession from mid-2006 to mid and in the recent years since the beginning of On the other hand, there are also several periods when the inverse relationship seems to be significant, especially, in the interval from mid until mid Considering the full sample, the correlation coefficient for the EU (-0.44) is negative. Table 2 summarizes the theory-consistency of consumers' expectations. When the period from January 1999 to May 2015 is considered, the EU (-0.44) and euro area (-0.53) correlation coefficients are close to the threshold value of As a result, euro-area consumers' expectations can be considered as consistent with the PC when the whole sample is utilized. Also German and Dutch consumers' expectations are theory-consistent. The rest of the observed MS shows correlation coefficients closer to zero. French, Spanish and UK coefficients are roughly equal to zero. The Italian correlation coefficient indicates weak negative correlation between consumers' expectations, while Polish consumers' expectations show weak positive correlation. When shorter periods of recessions (orange) and expansions (blue) are considered, the EU and euro area coefficients indicate very strong or almost perfect inverse relationship between the variables in the interval from May 2003 to May 7 Compare with Annex 1 and Annex 2. Annex 1 evaluates consumers' theory-consistency with the PC using consumers' expectations about future consumer prices developments (Q6) and expectations about development of general economic situation (Q4). Annex 2 replicates the analysis of section 3 and 4 by employing different variables. The change in the rate of unemployment is replaced by the rate of unemployment and a new variable question 7 cumulative is used instead of the original Q7. 8

9 2011. Therefore, consumers' expectations are theory-consistent in the periods covered by this interval, which encompasses two periods of expansion and one period of recession, the Great Recession. By contrast, during the latest recession (June 2011 to June 2014) the correlation coefficients are strongly positive, which conflicts with the PC. Table 2: Correlation coefficients for consumers' expectations - Q6 and Q7 Country 1999:1-2015: : : : :12 Period 2008: : : : : :06 EU EA Germany France Italy Spain Poland 0.14 N/A United Kingdom Netherlands At MS level, the values of correlation coefficients are mixed when the periods of expansions and recessions are considered. Similarly to the results using the hard data, there is no systematic pattern based on business cycle phases. During the first recession (July 2000 April 2003) and the first expansion (May 2003 December 2007) the results are rather mixed, with only a minority of selected MS showing theory-consistent consumers' expectations. By contrast, a majority of the MS shows significant theory-consistency in the period of the Great Recession (January 2008 to June 2009) and the following expansion (July 2009 to May 2011), whereas all correlation coefficients are positive or equal to zero in the latest recession (June 2011 to June 2014). In this period, consumers' expectations in none of the geographical areas indicate consistency with the PC. This period is represented by the simultaneous decline in inflation and unemployment expectations of consumers. All in all, within the selected MS, the Netherlands is the MS with the best theory-consistency of consumers' expectations. The value of the Dutch correlation coefficient for the full sample is the highest. Furthermore, all correlation coefficients except one are strongly negative when the shorter periods are considered. Comparing the theory-consistency of the hard data (Table 1, pp 7) and of consumers' expectations (Table 2), the following differences and similarities have been observed. Contrasting the full sample periods for the EU and the euro area, it is apparent that soft data indicate better theory-consistency than hard data, since their correlation coefficients show stronger negative correlation. 8 When looking at correlation coefficients of the EU and the euro area for the periods of expansions and recessions, consumers' expectations show better theory-consistency than the hard data, thanks to significantly 8 However, these results are slightly different when the rate of unemployment is used instead of the change in the unemployment rate (see Annex 2). 9

10 stronger negative correlation in the periods July 2000 to April 2003 and May 2003 to December The results in the other three periods are broadly the same. At MS level, for the whole period, consumers' expectations show somewhat better consistency with the PC than the hard data. None of the observed MS indicate theory-consistency using the hard data, while German and Dutch consumers' expectations are consistent with the PC according to the criteria used in this paper. When looking at periods of expansions and recessions, the difference between theory-consistency of the hard and soft data at MS level is not as pronounced as the difference observed at EU and euro-area level. Overall, strongly negative correlation coefficients are obtained in around 41% of the cases using soft data, while for the hard data this happens in 35% of the cases. Considering particular periods, the soft data indicate better theory-consistency for the first two periods (July 2000 to April 2003 and May 2003 to December 2007), while theory-consistency of hard data is slightly better in the last three analysed periods (January 2008 to June 2009, July 2009 to May 2011 and June 2011 to June 2014). 10

11 Inflation rate 5 Phillips curves for the analysed soft and hard data The objective of this section is to present the analysed hard and soft data in figures as Phillips curves. The periods of the business cycle are highlighted in the figures as well. Figures 5 and 6 chart the PCs for the hard and soft data. Figure 5 shows the relationship between the change in the rate of unemployment and the inflation rate (hard data) in the EU over the period from 2001 to 2014, plus the development in the first five months of This figure is complementary to Table 1 (pp 8). The only period in which a very clear and strong inverse relationship between the variables is visible is during the expansion of 2009 to The shape of the PC for the first two recessions (2001 to 2003 and 2007 to 2009) is equivocal as it shows neither a clear inverse relationship nor a positive linear relationship. However, this result can be attributable to the length of the analysed periods, which is rather short. Due to the very stable inflation rate of 2% in the expansion period 2003 to 2007, the PC is flat. Moreover, the PC for the latest recession (2011 to 2014) shows a positive linear relationship between the variables, which conflicts with the concept of the PC. Finally, the PC for the first five months of 2015 is practically vertical. Figure 5: EU Phillips curve inflation rate and change in the rate of unemployment ( , plus 2015 monthly data) 4% % 2% % % 2014 Apr Feb May Mar Jan % Change in the rate of unemployment Note: Red colour indicates recessions, blue colour indicates expansions. 11

12 Inflation expectations (Q6) Figure 6 charts the data of EU consumers' expectations on inflation (Q6) and unemployment (Q7) for the period from 2000 to 2014, plus the first six months of This figure is complementary to Table 2 (pp 10). The PC in Figure 6 shows - with few exceptions - almost a perfect continuous inverse relationship between consumers' expectations during the period from 2001 to On the contrary, the end of the observed period (2012 to 2014) indicates a positive linear relationship between the variables. However, the PC for the first six months of 2015 suggests again an inverse relationship for the soft data. Figure 6: EU Phillips curve Q6 and Q7 ( , plus 2015 monthly data) June May March April Feb Jan Unemployment expectations (Q7) Note: Red colour indicates recessions and blue colour indicates expansions. Comparing the PCs using hard or soft data (Figures 5 and 6), the most noticeable difference can be seen in the period 2003 to 2007 when the inflation rate in the EU remained stable while EU consumers' inflation expectations grew gradually throughout this period. 12

13 6 Theory-consistency according to consumers' categories The survey data provide the opportunity to evaluate consumers' expectations consistency with the PC at the level of the following consumer categories: gender, age, education and income. This section checks whether and how consumers' theory-consistency changes within the categories. Table 3 shows the difference in theory-consistency between male and female consumers. When the full sample is considered theory-consistency of men is somewhat better. On shorter periods, men's expectations are more theory-consistent in periods of recession, while women's expectations are as good as men's expectations or even slightly better during expansion periods. Table 3: Correlation coefficients for consumers' expectations gender categories Region EU EA Category 1999: : : : : :12 Period 2008: : : : : :06 Male Female Male Female Table 4 compares the differences in the correlation coefficients between four age groups of consumers. On the full sample period, inflation and unemployment expectations of younger respondents are noticeably more theory-consistent than expectations of older respondents: the younger the consumer, the better the theory-consistency of his or her expectations with the PC. This conclusion is valid for all shorter periods, except the Great Recession and the following expansion, when theory-consistency of all age groups is virtually the same in the former case and theory-consistency of the youngest age group is the worst in the latter case. Table 4: Correlation coefficients for consumers' expectations age categories Region EU EA Category 1999: : : : : :12 Period 2008: : : : : :06 AG AG AG AG AG AG AG AG

14 Table 5 presents the correlation coefficients of three groups of consumers with different levels of attained education. Consumers' theory-consistency does not change considerably with the level of consumers' education. On average, expectations of consumers with higher education are only slightly more theory-consistent, except during the period May 2003 to December 2007 where indeed correlation improves with higher education levels. Moreover, in the case of euro-area consumers, consumers with higher education score somewhat better than consumers with lower education. Surprisingly, in the period July 2000 to April 2003, euro-area consumers with lower educational level appear slightly more theory-consistent. Table 5: Correlation coefficients for consumers' expectations education categories Region EU EA Category 1999: : : : : :12 Period 2008: : : : : :06 ED ED ED ED ED ED Table 6 looks at the correlation coefficients of four income groups of consumers. Theory-consistency of consumers' expectations is improving with rising income: the higher the income, the stronger the negative correlation (or the weaker the positive correlation) between consumers' expectations. This conclusion, however, does not hold for the expansion of July 2009 to May 2011 when the correlation coefficients do not present any trend. Table 6: Correlation coefficients for consumers' expectations income categories Region EU EA Category 1999: : : : : :12 Period 2008: : : : : :06 RE RE RE RE RE RE RE RE

15 7 Conclusion The objective of this paper was to check whether European consumers' expectations about the future developments of inflation and unemployment are consistent with the Phillips curve. For this purpose, the Harmonised Consumer Survey questions Q6 and Q7 were used. The relationship between the two variables was evaluated using correlation coefficients, which were calculated for the time series of the aggregate soft data and eventually compared with the results for the respective hard data for the whole analysed period as well as for shorter periods of expansions and recessions. A first finding is that correlation coefficients obtained for the survey data indicate somewhat better theory-consistency than is observable for the hard data. For the whole analysed period, the analysis confirms the presence of an inverse relationship between consumers' expectations for both the EU and the euro area. However, the correlation coefficients are not always strong. At MS level, the results are mixed. German and Dutch consumers' expectations are clearly theory-consistent, while the rest of the observed MS shows correlation coefficients closer to zero. Similar results are observed at EU and euro-area level when consumers' expectations about the development of the general economic situation (instead of unemployment expectations) is employed to evaluate consumers' theory-consistency with the PC (Annex 1). Cumulating expectations of the change of the unemployment rate to obtain a measure of unemployment level expectations considerably weakens the results (Annex 2). When sub-periods are considered, the strength of the correlation coefficients varies significantly. At EU and euro-area level, an inverse relationship is present for all periods except one (June 2011 to June 2014 recession) when it changes into a strong positive linear relationship, which contradicts the PC. The differences in the values of correlation coefficients do not indicate any systemic influence based on development of the business cycle (changes in recession or expansion periods) on consumers' consistency with the PC. Using expectations about the general economic development, better theory-consistency of consumers is observed in the first and last analysed period. When Q7 cumulative is employed, the theoryconsistency of consumers is comparable to the results of the main analysis, with the exception of the last two sub-periods where correlation coefficients for most of the countries have different signs compared to the original analysis. Finally, the paper checks how consumers' theory consistency changes across different consumers' categories at EU and euro-area level. Generally, theory-consistency of men is better than theoryconsistency of women, younger generations are more theory-consistent than older generations, and consumers' consistency with the PC gets better with rising income, whereas higher education leads only to a marginal improvement in theory-consistency of consumers. In other words, a young rich man will, in general, be more theory-consistent than an elderly poor woman. Unlike Dräger, Lamla and Pfajfar (2013), which use microdata of the Michigan Survey of Consumers, the conclusions of this exploratory and descriptive note are based on the analysis of aggregate survey data. This does not allow checking what each individual replied to each question, but rather the average consistency of the total or a sub-group of the population. Therefore, the evidence presented in this paper is of a tentative nature and needs more research to be corroborated. In particular, an analysis of microdata from the Harmonised Consumer Survey might produce further insights. 15

16 References Dolan, Ed (2014): What Ever Happened to the Phillips Curve? Interpreting a Half Century of Inflation and Unemployment. Blog. Available at Dräger, Lena, Lamla, Michael J. and Pfajfar, Damjan (2013): Are Consumer Expectations Theory- Consistent? The Role of Macroeconomic Determinants and Central Bank Communication. KOF Working Papers, No European Commission (2014): The Joint Harmonised EU Programme of Business and Consumer Surveys. User Guide. Available at Friedman, Milton (1968): The Role of Monetary Policy. In: American Economic Review, Vol. 58, No. 1, Gordon, Robert J. (2011): The History of the Phillips Curve: Consensus and Bifurcation. In: Economica, Vol. 78, No. 309, Phelps, Edmund S. (1967): Phillips Curves, Expectations of Inflation and Optimal Employment over Time. In: Economica, Vol. 34, No. 3, Phillips, A. W. (1958): The Relation between Unemployment and the Rate of Change of Money Wage Rates in the United Kingdom, In: Economica, Vol. 25, No. 100,

17 Annex 1 Dräger, Lamla and Pfajfar (2013) in their analysis of consumers' expectations test also the relationship between inflation and business conditions, which they take as a proxy for the output gap. This approach is based on Okun's law and the New Keynesian PC. Okun's law is used to translate the trade-off between inflation and unemployment into a positive correlation between inflation and output, while the New Keynesian PC is defined as a function of expected inflation and the output gap. 9 The European Commission's Harmonised Consumer Survey contains the question 4 (Q4), which asks consumers on their expectations on development of general economic situation in their country. 10 Q4 can be applied together with question 6 (Q6) to test consumers' theory-consistency with the PC. In this case, consumers are consistent with the PC when the relationship between their expectations is characterized by a strong positive correlation. 11 Since the Q6 time series is lagging behind the Q4 time series, a lag of three quarters of year has been introduced to the Q4 data and the original Q4 time series has been replaced by the new time series: Q4(t-3). Figure 1 illustrates the relationship between the Q4(t-3) and Q6 from 1999 Q1 to 2015 Q1. Figure 1: Consumers' expectations in the EU Q4(t-3) and Q6 (quarterly data, Q to Q1 2015) EU Q6 EU Q4 (t-3) (rhs) r = Table 1 presents correlation coefficients, which were calculated for the full sample period (1999 Q1 to 2015 Q1). For the whole period, consumers in the EU are not strictly theory-consistent (positive correlation of 0.47), whereas consumers in the euro area show theory-consistency with the PC (positive correlation of 0.57). At MS level, most of the countries indicate a positive correlation but only three countries (Germany, Poland and the Netherlands) show theory-consistency with the PC. 9 This approach is formally described by Dräger, Lamla and Pfajfar (2013: 6). 10 The exact formulation of the question 4 can be found in the survey user guide, which is available at 11 For this analysis quarterly survey data have been employed. 17

18 When shorter periods are considered, EU and euro-area consumers are theory-consistent in all recessions and expansions, except the latest recession of In the recession of , consumers' expectations show almost a perfect correlation. At MS level, the results are mixed but negative correlation is present only in very few cases. Table 1: Correlation coefficients for Q4(t-3) and Q6 (quarterly data) Country 1999:1-2015:1 2000:3-2003:1 2003:2-2007:4 Period 2008:1-2009:2 2009:3-2011:2 2011:3-2014:2 EU EA Germany France Italy Spain Poland 0.61 N/A United Kingdom Netherlands

19 Annex 2 Theory-consistency of hard data is different when the rate of unemployment is utilized instead of the change in unemployment rate. The analysis, which uses the rate of unemployment is viable because the PC describes the relationship between the rate of unemployment and the rate of inflation. Therefore such an analysis matches the concept better. Figure 1 illustrates the relationship between the EU inflation rate and the EU rate of unemployment for the period January 2000 to May 2015, where the correlation is negative but weak (-0.43). Figure 1: The EU inflation rate and the EU rate of unemployment (January 2000 to May 2015) 5.0% 4.0% r = % 11.0% 3.0% 2.0% 1.0% 0.0% -1.0% EU Inflation rate EU Rate of unemployment (rhs) 10.0% 9.0% 8.0% 7.0% 6.0% 5.0% Table 1 shows the correlation coefficients for the inflation rate and the rate of unemployment in the EU, euro area and seven selected MS. When the full sample is applied the euro area, France, Spain and the Netherlands indicate theory-consistency with the PC. The EU and Italy are very close to the threshold correlation coefficient of Germany shows no correlation, whereas Poland shows a weak positive correlation and the UK shows even a strong positive correlation between inflation and unemployment. The Netherlands is the country with the most theory-consistent data. In all three recessions the Dutch correlation coefficients reached the level of strong negative correlation. The EU and the euro area reached such a level of strong negative correlation only twice during recessions. All the other countries, except the UK, recorded two shorter periods with a strong negative correlation. The EU and the euro area values of correlation coefficients for the given periods are in several cases very close but in other cases they differ significantly. During the expansions (May 2003 to December 2007 and July 2009 to May 2011), the hard data for the EU and the euro area do not show consistency with the PC. The results are much more positive (theory-consistent) for the recession periods (July 2000 to April 2003, January 2008 to June 2009, June 2011 to June 2014) when at least one of the European regions indicates negative and strong correlation between the inflation rate and the rate of unemployment. In those cases the hard data are consistent with the PC. In general, correlation between inflation and unemployment tends to be stronger and negative in the periods of 19

20 recession in comparison to the periods of expansion. This observation has to be considered carefully because the examined period is rather short and might not be representative. Table 1: Correlation coefficients for the inflation rate and the rate of unemployment Country 2000: : : : : :12 Period 2008: : : : : :06 EU EA Germany France Italy Spain Poland United Kingdom Netherlands In addition to the original Q7, which does not comply with the concept of the PC completely, a new variable question 7 cumulative is introduced. Indeed, the Q7 does not ask the consumers for their expectations of the future level of unemployment, but the change in the number of unemployed people, which is not exactly the same variable as the rate of unemployment. Therefore a transformation of the original time series is carried out. First, the series is standardized and then the standardized values are cumulated 12. As a result of this transformation the new variable question 7 cumulative expresses the level of unemployment based on consumers' expectations. It is comparable with the rate of unemployment. Figure 2 depicts the question 7 cumulative for the EU data with the EU rate of unemployment. These two time series are highly correlated (0.91). The transformed time series of the question 7 correlates almost perfectly with the unemployment rate. From Figure 2, it is apparent that the series of the question 7 cumulative lags behind the rate of unemployment. 12 More precisely, the cumulative Q7 series (Q7*) is calculated in the following way: DMQ7 n = Q7 n long-term average(q7) Q7 n = Q7 n 1 + DMQ7 n; where n Z +, Q7 0 = DMQ

21 Figure 2: EU Q7 cumulative and the EU unemployment rate (January 2000 to May 2015) EU unemployment rate EU Q7 cumulative (rhs) r = Figure 3 presents the time series of EU consumers' expectations on inflation (Q6) and cumulative consumers' unemployment expectations (Q7 cumulative), which corresponds with the rate of unemployment. For the whole period January 1999 to May 2015, the correlation between these two variables is negative but weak (-0.13). Figure 3: Consumers' expectations in the EU - Q6 and Q7 cumulative (January 1999 to May 2015) EU Q6 EU Q7 cumulative (rhs) r = Table 2 shows the correlation coefficients, which were calculated for the Q6 and the Q7 cumulative. For the whole period, consumers' expectations are not theory-consistent. The correlation coefficient for the EU is very weak (-0.13) and the correlation coefficient for the euro area is even weaker (-0.09). For the full sample, at MS level, only Spain shows strong negative correlation and thus theory-consistency. The rest of MS indicates rather weak correlation (France, Poland and the Netherlands), no correlation (Germany) or even positive correlation (the UK, Italy). 21

22 Table 2: Correlation coefficients for consumers' expectations - Q6 and Q7 cumulative Country 1999: : : :04 Period - soft data - Q7_cum 2003: : : : : : : :06 EU EA Germany France Italy Spain Poland N/A United Kingdom Netherlands When the business cycle phases are considered, the results are mixed at all levels. Theoryconsistency of consumers' expectations is not systematically influenced by development of the business cycle. EU and euro-area consumers' expectations are theory-consistent in three out of five periods (May 2003 to December 2012, January 2008 to June 2009 and June 2011 to June 2014). For the other two periods (July 2000 to April 2003 and July 2009 to May 2011) correlation coefficients are positive. At MS level, four MS (France, Italy, Spain and the Netherlands) show theoryconsistency in three periods, whereas the other two MS (Germany and Poland) show only one period of theory-consistency. The period July 2009 to May 2011 is dominated by very strong positive correlation coefficients in most of the observed MS. For a comparison on the whole period, consumers' theory-consistency in the EU and euro area is substantially stronger when the original Q7 is employed. At MS level, the difference is not that strong. When the periods of recessions and expansions are considered, the most significant difference at all levels is present in the last two observed periods where the correlations go in opposite directions. 22