Online collection and return

1
OECD Workshop on Business and Consumer Tendency
Surveys Rome, 19 September 2006
Long-Run Biases in Consumer Sentiment
Micro Evidence from European Surveys
Maurizio Bovi ISAE, Italy
2
Plan

• Fit
• Goal and Contribution
• Motivations
• Data
• Statistical Framework
• Experiments and Results
• Puzzles or well known Psycho-Biases?
• Concluding Remarks

3
Fit

• Data users often take consumer sentiment indexes
  (CSI) as a given input and, by and large, they
  search for links between CSI and Economic System
  hard data.
• Data producers think about CSI as a final output,
  addressing issues such as data collection,
  response rates, etc. Roughly speaking, they deal
  with mapping Individuals into CSI.
• My research examines semi-worked survey data
  without reference to hard data and focuses on
  their reliability and on how Individuals address
  the Economic System.

4
Goal and Contribution

• A long-run analysis of the consumer sentiment,
  taking advantage of micro data ( of respondents)
  and cognitive psychology findings.
• Keywords
• long-run analysis
• representative consumer
• micro data.

5
Motivations

• Why micro data?
• In many political and economic circles, CSI are
  commonly diffused, commented and studied at their
  face value. Looking at CSI components may be
  interesting even from the data users point of
  view.
• Analyzing micro data allows avoiding comparisons
  with National Account data which, in turn,
  reduces problems such as the vagueness/difficulty
  of the queries. E.g., how does the respondent
  interpret queries about general economic
  conditions?

6
Motivations (2)

• Why a long-run analysis?
• Inter alia, this Workshop is about consumer
  tendency.
• I have enough data (assuming that twenty years
  are enough for a long-run analysis).
• To some extent, a long-run approach may sidestep
  some data issues (changes in survey methods,
  seasonality, etc.)

7
Data
Data are from the Business Surveys Unit of the
European Commission. I deal with the following
queries and with the relative reply options Q1)
How has the financial situation of your household
changed over the last 12 months? It has ... Q2)
How do you expect the financial position of your
household to change over the next 12
months? It will ... Q3) How do you think the
general economic situation in the country has
changed over the past 12 months? It has ... Q4)
How do you expect the general economic situation
in the country to develop over the next 12
months? It will ... PP) get/got a lot
better P) get/got a little better E)
stay/stayed the same M) get/got a
little worse MM) get/got a lot worse N)
don't know.
8
Data (2)

• are percentages of respondents MMMEPPPN100
• refer to fifteen European Union (EU) countries
• start in January 1985 for nine out of fifteen
  countries
• Exemptions are
• Austria (starting date 199510),
• Finland (starting date 198711),
• Luxembourg (starting date 200201),
• Portugal (starting date 198606),
• Spain (starting date 198606),
• Sweden (starting date 199510).
• stop in July 2005 for all countries.

9
Statistical Framework

• I analyze
• stylized facts via
• full-sample descriptive statistics about
• representative consumers within
• the survey framework
• All that should reduce some data issues (the lack
  of re-interviews, changes in survey methods,
  seasonality, vagueness of the queries), allowing
  robust findings.

10
Experiments and Results

• Some simple experiments based on reply options
  allow verifying the persistent presence of
  logical behaviors. They may be thought of as
  somewhat supporting the reliability of the
  answers. For instance
• Consumers should know their own situation better
  than the system wide one.
• Thus, e.g., the average share of individuals
  answering dont know to questions about the
  general economic environment should be greater
  than the average share of individuals which do
  not know how their own financial situation is
  going on.
• Consumers should know past situations better than
  future ones.
• Thus, ex ante questions should show more dont
  know than the corresponding ex post ones.

11
Consumers Uncertainty on Personal vs General and
on Past vs Future Economic Conditions
EU_11Belgium, Germany, Denmark, Greece, Spain,
Finland, France, Ireland, Italy, Netherlands, UK
(sample 8711-0507). Full sample average of
responses dont know (in of total) to the
questions Q1-Q4.
12
Experiments and Results (2)

• Let us now turn the attention to the E answer.
  There are reasons to believe that it should show
  the largest scores
• Since the queries are about dynamics,
  individuals should respond, on average, the
  same the most part of times, because it is hard
  to think of ever improving/worsening economic
  conditions over many years. It is important to
  note that it should hold whatever economic
  conditions means for common people.
• The preference of being E may be partly due to
  short-cut heuristics - this neutral option may
  be chosen by uninformed and/or uninterested
  respondents.

13
Fig. 1. Distribution of Responses On Economic
Conditions
Histograms report full sample (8711-0507) means
of each reply item. EU11Belgium, Germany,
Denmark, Greece, Spain, Finland, France, Ireland,
Italy, Netherlands, UK.
14
Experiments and Results (3)

• Given their logically expected outcomes, all
  the tests performed so far on N and E lead to
  think that data give a faithful representation of
  peoples opinions. In passing, neither N nor E
  enter into CSI.
• Data tell more than this. Figure 1 shows that the
  number of E-agents is structurally much higher
  when the question is about personal (Q1, Q2) as
  opposed to general (Q3, Q4) economic developments
  (more than 55 vs less than 40).
• This calls for ad hoc tests to contrast general
  vs personal response options. One way to proceed
  is computing mean values of (Q1Q2)-(Q3Q4) for
  each single option. When referring to MM and
  M (PP and P) percentages, negative
  (positive) values imply that the personal
  condition is perceived to be systematically
  better than the general one.

15
Personal vs General Sentiment in European
Countries

EU_11Belgium, Germany, Denmark, Greece, Spain,
Finland, France, Ireland, Italy, Netherlands, UK
(sample 8711-0507). Black values indicate
that the personal condition is perceived to be
better than the general one.
16
Experiments and Results (4)

• According to one of the basic axiom of standard
  neoclassical models, agents should not persist in
  repeating the same mistake.
• In the present framework, it may be addressed by
  looking at the gap between contemporaneous ex
  ante (Q2, Q4) and ex post (Q1, Q3) responses, to
  which I refer as the forecast error (iPP, P,
  E, M, MM)
• forecast error 100Q1i-Q2i-12)/Q1iQ2i-12
• Likewise for general conditions (Q3-Q4).
• For MM and M, positive errors indicate
  over-optimistic expectations and/or
  over-pessimistic judgments. E.g., today 30
  judges the last a worse year, 12 months ago 10
  foresaw it as a worse year. The reverse holds
  for PP and P.
• It is noteworthy that, in this setting, there is
  no need for agents to correctly address what an
  economic situation really is. In fact, I just
  compare answers given to the same question.

17
Europeans Forecast Errors (MM)
18
Europeans Forecast Errors (MM)
19
Europeans Forecast Errors Statistics(mean in
5 band)
(PP)
(P)
(E)
20
Europeans Forecast Errors Statistics(mean in
5 band)
(M)
(MM)
21
Puzzling Results

• Peoples tendency to judge over-pessimistically
  and/or to forecast over-optimistically.
• The ambiguity arises because of the lack of a
  hard benchmark (e.g., GDP, Consumption, etc.).
  However, it implies that
• peoples forecasts show a long run bias.
• peoples tendency to think that their own
  economic situation is better than the general one
  - the representative consumer think to become
  richer than himself.
• To sum up, there seems to be a structural mantra
  echoing across Europe
• As Usual, it Has Got Worse Than I Expected.
• Especially for the Others.
• Nevertheless, I Still Think That it Will Get
  Better.
• Especially for Me.

22
Puzzles or well known Psycho-Biases?

• Over-pessimism in judgments
• Availability bias. Mere repetition of certain
  information in the media, regardless of its
  accuracy, makes it more easily available and
  therefore falsely perceived as more accurate.
  Since the media tend to overweight bad economic
  news (Doms and Morin, 2004), there are reasons
  inducing individuals toward dispositional
  pessimism. Moreover, the information flow may
  also run from people to media (Curtin, 2003),
  creating a perverse spiral.
• Over-optimism in forecasts
• Irrational exuberance. In uncertain situations
  people tend to make forecasts by assuming, often
  without sufficient reasoning, that future
  favorable patterns will resemble past ones.
• Law of small numbers. People tend to
  over-inference from too short sequences of
  observations.
• Hindsight/Confirmation bias. Individuals tend to
  concoct ex post logical explanations for ex
  ante totally unexpected events.
• All that prevents agents from adequately learning
  from the past and from being aware of their
  errors, leading to long-run biases.

23
Puzzles or well known Psycho-Biases? (2)

• Over-pessimism in judgments and
• over-optimism in forecasts
• Mental Accounting. People allocate current and
  future income in different accounts.
• Over-self-confidence
• Illusion of Control. People have an expectancy of
  a personal success probability inappropriately
  higher than the objective probability would
  warrant.
• Depressive realism. One interpretation of it is
  that non-depressed people possess a positive
  bias, which allows them to feel in control of
  their environment. Since, hopefully, the
  representative European is non-depressed,
  evidence supports the agents egocentric bias.

24
Concluding remarks

• When elicited about economic conditions, people
  tend to reply both as expected and irrationally.
• Empirical evidence highlights paradoxical
  (rectius, emotionally-driven) responses, even
  when dealing with familiar conditions.
• Thus, it is not only a problem of the
  amount/quality of available information and/or
  the difficulty of the exercise - there is
  something else preventing Muthian results.
• My research suggests that psychology may be of
  some help. To the extent it is true
• not necessarily the detected puzzling outcomes
  reduce the reliability of survey data,
• it may be useful adding psychological
  considerations to CSI biases are structural gt
  manageable

25
After all, everybody should agree that the
sentiment is a mix of rationality and feelings