Consumers quantitative inflation perceptions and expectations provisional results from a joint study Rodolfo Arioli, Colm Bates, Heinz Dieden, Aidan Meyler and Iskra Pavlova (ECB) Roberta Friz and Christian Gayer (DG-ECFIN)
Outline 1. Motivation. The EC consumer survey 3. Empirical features of the dataset Aggregate and national results Inflation assessments across different groups Distributional characteristics Cross-checking qualitative and quantitative Business cycle effects 4. Addressing bias and extracting more information. Next steps
1. Motivation Review/update/extend earlier studies (1, 6) Country, euro area and EU results Economic situation (post-crisis, low inflation environment) Large micro data set allows for numerous analytical aspects 3
. The EC Consumer survey Qualitative questions How do you think that consumer prices have developed over the last 1 months? They have: [1] risen a lot [4] stayed about the same [] risen moderately [] fallen [3] risen slightly [6] don t know By comparison with the past 1 months, how do you expect that consumer prices will develop over the next 1 months? They will: [1] increase more rapidly [4] stay about the same [] increase at the same rate [] fall [3] increase at a slower rate [6] don t know 4
. The EC Consumer survey Qualitative questions (balances) and HICP (annual percentage changes)
. The EC Consumer survey Qualitative (balances) and quantitative (annual percentage changes) questions EU Euro area 6
. The EC Consumer survey Two approaches for aggregating European totals: Independent country distributions means and standard deviations for the total sample and for socio-economic breakdowns EU/euro area distribution higher moment statistics trimming 7
3. Empirical features Quantitative inflation estimates: Descriptive results Overestimation and bias Diversity (socio-demographics, national data) Distribution, trimming, functional forms Cross checking qualitative and quantitative Business cycle effects 8
4 6 7 8 9 1 11 1 13 14 1 4 6 7 8 9 1 11 1 13 14 1 3.1 Aggregate results Broad co-movement of both quantified inflation perceptions and expectations with actual inflation But systematic bias remains a feature EU EA Inflation perceptions Inflation expectations HICP Inflation perceptions Inflation expectations HICP 1 1 1 1 - - 9
-1-11 -1-9 -8-7 -6 - -4-3 - -11-1 -11-1 -9-8 -7-6 - -4-3 - -11 11 1 11 1-1 -11-1 -9-8 -7-6 - -4-3 - -11 3.1 Correlation with actual inflation -1-11 -1-9 -8-7 -6 - -4-3 - -11 11 1 11 1-1 -11-1 -9-8 -7-6 - -4-3 - -11 1..8.6.4.. -. -.4 EA 34 67-1 -11-1 -9-8 -7-6 - -4-3 - -11 8 9 1 11 1 11 1 Perceived and actual inflation Expected and actual inflation Perceived and expected inflation 1..8.6.4.. -. -.4 EA 34 67 8 9 1 1..8.6.4.. -. -.4 EA 34 67 8 9 1 1..8.6.4.. -. -.4 EU 34 67 8 9 1 1..8.6.4.. -. -.4 EU 34 67 8 9 1 1..8.6.4.. -. -.4 EU 34 67 8 9 1 Perceived inflation slightly (1- months) lags actual inflation Expected inflation a little more contemporaneous (-1 months) with actual inflation but clearly lags target outcome (1 months ahead) Kink in correlation structure between perceived and expected inflation 1
4 6 7 8 9 1 11 1 13 14 1 4 6 7 8 9 1 11 1 13 14 1 3.1 Bias has varied over time Both for perceived and expected inflation but particularly the former 1 1 EU Q1 minus EU HICP EA Q1 minus EA HICP 1 1 EU Q61 minus EU HICP EA Q61 minus EA HICP 11
4 6 7 8 9 1 11 1 13 14 1 4 6 7 8 9 1 11 1 13 14 1 4 6 7 8 9 1 11 1 13 14 1 4 6 7 8 9 1 11 1 13 14 1 3.1 Differences across countries Biases less for Nordic countries and FR, higher for IT, ES and EL (also DE post changeover) 4 3 3 1 1 - DK FI SE FR DK FI SE FR 1 1-4 3 3 1 1 - DE EL ES IT DE EL ES IT 1 1-1
4 6 3.1 Empirical features (cont d) Survey design and bias UK surveys of perceptions and expectations have less bias GfK-BOE quarterly survey and You-Gov Citigroup monthly survey Questions focus the respondent on ranges Jan- Jan- Jan-1 Jan-1 Time UK HICP YouGov-Citigroup perceptions GFK expectations GFK perceptions 13
- 1 1 3.1 Empirical features (cont d) Survey design and bias US inflation expectations and CPI 198 198 199 199 1 1 date Inflation Expectation Urban Area CPI (sa) US Michigan Survey long established and relative little bias The Stay about the same answers probed for meaning (levels or rates) Answers above % are probed Don t know answers are asked about cents on the dollar Sample relatively small c.a. individuals per month 14
1 1 mean inflation perception and expectation 1 1 3. Different groups, different assessments Different people, different inflation assessments? Opinions differ across different socio-demographic groups Men, older respondents tend to give lower (more accurate) answers with respect to their counterparts Inflation perceptions and expectation tend to decrease with level of income and education attainment Mean inflation perception and expectation by gender and age Mean inflation perception and expectation by income and education male female male female male female male female 16-9 3-49 -64 6+ mean inflation perception mean inflation expectation 1st nd 3rd 4th 1st nd 3rd 4th 1st nd 3rd 4th primary secondary further mean inflation perception mean inflation expectation 1
- - 4 Mean inflation perception and expectation 4 - - 4 Mean inflation perception and expectation 4 3. Different groups, different assessments Different people, different inflation assessments? The distribution of answers by men, high income earners and respondents with high level of education is less skewed and narrower Inflation perception and expectation by gender Inflation perception and expectation by age male excludes outside values mean inflation perception female mean inflation expectation 16-9 3-49 -64 6+ mean inflation perception mean inflation expectation excludes outside values Inflation perception and expectation by level of education Inflation perception and expectation by income primary secondary further mean inflation perception mean inflation expectation excludes outside values 1st quartile nd quartile 3rd quartile 4th quartile mean inflation perception mean inflation expectation excludes outside values 16
Percent Percent Percent Percent 3. Different groups, different assessments Different people, different inflation assessments? Women, older respondents, less educated people and those with lower income are less inclined to provide a quantitative answer 1% Share of quantitative replies on inflation perception by gender 1% Share of quantitative replies on inflation perception by age 8% 8% 6% 6% 4% 4% % % % male No answer female Answer % 16-9 3-49 -6 6+ No answer Answer 1% Share of quantitative replies on inflation perception by level of education 1% Share of quantitative replies on inflation perception by income 8% 8% 6% 6% 4% 4% % % % primary secondary further % 1st quartile nd quartile 3rd quartile 4th quartile no reply No answer Answer No answer Answer 17
. Density 1 1. 3.3 Distribution of replies - histogram Some extreme outliers limited in number histogram q1a width(.1) - quantitative perceptions - 1 q1 with negative values 18
.. Density 1 Density 1 1. 1. 3.3 Distribution of replies - histogram Zooming in truncated at ; peaks @, 1 histogram q1a width(.1) - quantitative perceptions histogram q1a ifq1a >=- & ifq1a <=6, width(.1) - quantitative perceptions - 1 q1 with negative values - 4 6 q1 with negative values 19
. Density 1 1... Density 1 Density 1 1. 1. 3.3 Distribution of replies - histogram Zooming in * mini-peaks at 1,, 3 histogram q1a width(.1) - quantitative perceptions histogram q1a ifq1a >=- & ifq1a <=6, width(.1) - quantitative perceptions - 1 q1 with negative values - 4 6 q1 with negative values histogram q1a ifq1a >=- & ifq1a <=3, width(.1) - quantitative perceptions -1 1 3 q1 with negative values
. Density 1 Density 1. 1 3.. Density 1 Density 1 1. 1. 3.3 Distribution of replies - histogram Full zoom and 3 modal replies of mini peaks histogram q1a width(.1) - quantitative perceptions histogram q1a ifq1a >=- & ifq1a <=6, width(.1) - quantitative perceptions - 1 q1 with negative values - 4 6 q1 with negative values histogram q1a ifq1a >=- & ifq1a <=3, width(.1) - quantitative perceptions histogram q1a ifq1a >= & ifq1a <=1, width(.1) - quantitative perceptions -1 1 3 q1 with negative values 1 4 6 8 1 q1 with negative values
4m1 m1 6m1 7m1 8m1 9m1 1m1 11m1 1m1 13m1 14m1 1m1 4m1 m1 6m1 7m1 8m1 9m1 1m1 11m1 1m1 13m1 14m1 1m1 3.3 Other features of distribution over time Median sometimes uninformative owing to clustering Whilst minima and maxima are extreme, 1 th and 9 th percentiles are less so Standard deviation has declined over time (as has interquartile range) Skew and kurtosis are persistent features 4 4 3 3 1 1 - mean p1(q1a) p(q1a) med(q1a) p7(q1a) p9(q1a) inflation 4 4 3 3 1 1 - mean p1(q61a) p(q61a) med(q61a) p7(q61a) p9(q61a) inflation
. 1 Density 1. 1 3 4 3.3 But histogram not uninformative Although some features prevail (peaks @, 1, etc., truncation at, etc.), there is a clear shift in distribution between June 8 (HICP = 4.%) and Jan 1 (HICP = -.6%) histogram q1a ifq1a >=- & ifq1a <=3, width(.1) - quantitative perceptions histogram q1a ifq1a >=- & ifq1a <=3, width(.1) - quantitative perceptions -1 1 3 q1 with negative values -1 1 3 q1 with negative values 3
3.4 What do slightly, moderately, a lot imply? Another advantage of the quantified data is that it allows us to consider whether consumers qualitative assessments are state/time dependent In aggregate they are internally consistent 4
3.4 I say slightly, you say moderately? There is some overlap between respondents definitions of a lot, moderately and slightly mean (pp) p(q1a pp) p7(q1a pp) mean (p) p(q1a p) p7(q1a p) 6 4 3 1 3 1 1 1 1 8 6 4 mean (s) p(q1a s) p7(q1a s) -1 - -3-4 - -6 mean p(q1a) p7(q1a) RTD check how much owes to inter vs intra country
4m1 m1 6m1 7m1 8m1 9m1 1m1 11m1 1m1 13m1 14m1 1m1 4m1 m1 6m1 7m1 8m1 9m1 1m1 11m1 1m1 13m1 14m1 1m1 4m1 m1 6m1 7m1 8m1 9m1 1m1 11m1 1m1 13m1 14m1 1m1 4m1 m1 6m1 7m1 8m1 9m1 1m1 11m1 1m1 13m1 14m1 1m1 4m1 m1 6m1 7m1 8m1 9m1 1m1 11m1 1m1 13m1 14m1 1m1 3.4 Changes driven by assessment not definition Evolution of inflation perceptions not driven by changing definition of slightly, moderately, etc. but rather by changing assessments of inflation 8, 7, 6,, 4, 3,, 1, N (pp) hicp 4 3 1-1,, 4, 4, 3, 3,, N (p) hicp 4 3 1-1, 4, 3,, 1, N (s) hicp -1 1 3 4 N (n) hicp N (nn) hicp 8, 6, 4,, -1 1 3 4 1,4 1, 1, 8 6 4-1 1 3 4 Note: pp denotes a lot, p ~ moderately ; s ~ slightly ; n ~ same ; nn ~ fallen NB: HICP inflation scaling inverted for s, n and nn 6
3.. Correlation with real economy Consider whether inflation assessment is linked to business cycle Tentative evidence of limited effect bias slightly higher during recession (causality?) Since 9, bias fractionally higher with low inflation owing to downward rigidity? Table Consumers quantitative estimates of inflation and HICP euro area (annual percentage changes; Jan 4 Jul 1) Q1 Q61 HICP difference Q1 - HICP difference Q61 - HICP 3: 7:1 expansion 1.4 6.. 1.3 4.3 8:1 9:6 recession 1.8 7..4 1.4 4.6 9:7 11:9 expansion.9 3.8 1.6 4.3. 11:1 13:3 recession 8.1.4..6 3. 13:4 Jul 1.8 3.6.6. 3.1 Table 6 Consumers quantitative estimates of inflation and HICP euro area (annual percentage changes; Jan 4 Jul 1) Q1 Q61 HICP difference Q1 - HICP difference Q61 - HICP Jan 4 - Feb 9 above 1% 1.9 6.9.3 1. 4. Mar 9 - Feb 1 below or equal 1%.6.9..3.7 Mar 1 - Sep 13 above 1% 7. 4.8...6 Oct 13 - Jul 1 below or equal 1%.4 3.4.3.1 3.1 7
3.. Correlation with real economy Some evidence that inflation assessment lags real economy, but much of this is driven by crisis period. However this is not so robust nor stable - since 1 correlation has changed sign Chart X.a: Leading and lagging correlations Euro Area (percentage points) 8
4m1 m1 6m1 7m1 8m1 9m1 1m1 11m1 1m1 13m1 14m1 1m1 4.1 Signal extraction trimming Previous analyses (6, 1) also considered trimming A little (e.g. 1% or 3%) goes a long way Thereafter, diminishing returns to scale "HICP inflation" "Q1" "Q1t1s1" "Q1t3s1" "Q1ts1" "Q1t68s1" "Q1t9s1" 1 1-9
4m1 m1 6m1 7m1 8m1 9m1 1m1 11m1 1m1 13m1 14m1 1m1 4.1 Signal extraction trimming with asymmetry Previous analyses considered trimming but not asymmetric Asymmetry reduces (eliminates) bias (post crisis period) "HICP inflation" "Q1" "Q1ts1" "Q1ts1" "Q1ts17" 1 1-3
4m1 m1 6m1 7m1 8m1 9m1 1m1 11m1 1m1 13m1 14m1 1m1 4.1 Signal extraction trimming with asymmetry Asymmetry reduces (eliminates) bias (post crisis period) But cannot handle pre- and post-crisis periods "HICP inflation" "Q1" "Q1t68s1" "Q1t68s1" "Q1t68s17" 1 1-31
4. Signal extraction log-normal distribution Two features of data truncation at zero and long upper tail suggest log-normal distribution might fit well mean* EU HICP mode* EU HICP 16 14 1 1 8 6 4-8 6 4 - -4 location EU HICP scale EU HICP 3.1 3..9.8.7.6. 4. 4. 3. 3... 1. 1... -. -1..8.8.7.7.6.6...4-1. -... 1. 1... 3. 3. 4. 4. Log-normal distribution captures the mode/peak of replies and is closer to the actual outcome than the mean, which is excessively influenced by large positive outliers 3
<-1-1 1 3 3 4 4 <-1-7 -4-1 8 11 14 17 3 6 9 3 3 38 41 44 47 4.3 Round Numbers Suggest Round Interpretation? Binder (1) using Michigan data argues that more uncertain respondents are more likely to report multiples of five 1 1 actual multiple of 1 multiple of multiple of 1 1 1 actual rounded to 1 multiple of multiple of 1 In EU case, actual distribution of responses can be replicated using mix of three low, middle and high uncertainty distributions Although yet to be fully implemented could enable information from full distribution to be exploited 33
.1 First results and initial assessment Updated results confirm earlier findings Bias but broad co-movement with inflation Dist. - upward skew and strongly kurtotic Systematic (but limited) differences across groupings National differences remain source? Impact of the crisis seems relatively limited Controlling for distributional features, it may be possible to extract credible signal for policy makers 34
. Next steps Complete preliminary report Research agenda Implement Binder (1) approach Exploit other ways to screen/cleanse the data e.g. check consistency of replies with Philips Curve, à la Dräger/Lamla/Pfajfar (13) Consider link between inflation assessment and other survey variables (confidence, buying intentions, saving intentions, etc.) 3
Consumers quantitative inflation perceptions and expectations provisional results from a joint study Rodolfo Arioli, Colm Bates, Heinz Dieden, Aidan Meyler and Iskra Pavlova (ECB) Roberta Friz and Christian Gayer (DG-ECFIN)