Acquisition of information and share prices : An empirical investigation of cognitive dissonance

Acquisition of information and share prices: An
empirical investigation of cognitive dissonance?
Elena Argentesi†
Helmut L¨
utkepohl‡
Massimo Motta§
This draft: 19 March 2009
First draft: 23 March 2005
Abstract
This paper deals with the determinants of agents’ acquisition of informa-
tion. Our econometric evidence shows that the general index of Italian share-
prices and the series of Italy’s ?nancial newspaper sales are cointegrated, and
the former series Granger-causes the latter, thereby giving support to the
cognitive dissonance hypothesis: (non-professional) agents tend to buy the
newspaper when share prices are high and not to buy it when share prices
are low. Instead, we do not ?nd support for the hypothesis that the agents
acquire information in order to trade in the stock-market: we ?nd no rela-
tionship between quantities exchanged in the market and newspaper sales,
nor between stock market volatility and newspaper sales.
Keywords: Cointegration, cognitive dissonance, Granger-causality, ?nancial news-
paper sales, stock market and information
JEL classi?cation: C32, D03, G14
?We are grateful to Armando Carcaterra, Claudio Mazzoli (Brainpower) and Hammad Nasar for
their help with Italian stock-market data, to Sibilla Guzzetti (ADS) for help with the newspapers
sales data, and to Andrea Beltratti, Giuseppe Bertola, Laura Bottazzi, Antonio Cabrales, Juan
Carrillo, Stefano DellaVigna, Luigi Guiso, Elisabetta Iossa, Tullio Jappelli, Bruno Jullien, Gregor
Langus, Annamaria Lusardi, Marco Pagano, Ludovic Renou, Riccardo Rovelli, Karl Schlag, and
seminar participants at the EUI (Florence), Bologna, WZB (Berlin), HECER (Helsinki), ESEM
2006 (Vienna) and Ausschuss f¨
ur ¨
Okonometrie 2008 (Rauischholzhausen) for useful discussions.
Moreover, we are grateful to Walter Kr¨amer, the Editor, and two referees for comments which
helped us to improve the presentation of the results.
†University of Bologna. Address: Department of Economics, Piazza Scaravilli 2, I-40126
Bologna.
‡European University Institute, Florence. Address: EUI, Department of Economics, Via della
Piazzuola 43, I-50133 Firenze, email: helmut.luetkepohl@eui.eu.
§University of Bologna and CEPR. Address: Department of Economics, Piazza Scaravilli 2,
I-40126 Bologna.

---

1
Introduction
In ‘standard’ economic theory, an agent will normally be better o? by having more
information, if the latter was free. This is because utility depends on outcomes, and
information (if it has any relevance) should help an agent take better decisions, in
turn improving outcomes. In recent years, however, a growing literature has inte-
grated psychology into economics, suggesting various reasons why agents might want
not to acquire available information. Di?erent ways to model this phenomenon have
been proposed, including strategic behavior by agents, as well as the incorporation
of beliefs in the utility function of individuals (see below for references).
To investigate empirically whether information acquisition is driven by psycho-
logical considerations, we look at the relationship between non-professional investors’
acquisition of information about ?nancial markets - in the particular form of their
purchase of Italy’s main ?nancial newspaper, Il Sole 24Ore - and data on the evo-
lution of the Italian stock exchange market.
Our hypothesis is that this relationship is consistent with the theory of cognitive
dissonance (Festinger, 1957), very in?uential in social psychology, and supported
by a number of anecdotal and experimental ?ndings, in very di?erent domains and
contexts. This theory postulates that two cognitions (or elements of knowledge) are
dissonant if the opposite of one cognition follows from the other. Dissonance makes
an agent uncomfortable, and in order to reduce it the agent may either avoid any
information likely to create dissonance or process the available information so as to
reduce it. In our context, an agent who learns that the price of the shares she holds
decreases will experience dissonance: the cognition that a share is ‘doing badly’ is
dissonant with the cognition that she holds that share in her portfolio. In order
to reduce dissonance the agent can keep her behavior unchanged (i.e., continue to
hold the share) while eliminating the dissonant knowledge by ignoring information
about the share price. Therefore, at an aggregate level this theory suggests that
agents acquire information when the share-price index increases (that is, they buy
the ?nancial newspaper when they expect to see that the particular shares they
hold are doing well), but prefer to ignore information when the share-price index
decreases (that is, they do not buy the newspaper when they expect to learn from
1

---

it that their shares are not doing well). We assume that expectations are correct on
average, as the agent is exposed to some rumor about the general price level, but
it is only after buying a newspaper that she will have precise information on her
shares. Another way to formalize this idea is that, if share prices are correlated over
time (as it is argued by the ?nancial literature on ‘mean reversion’, see for instance
Fama and French, 1988), then the agent may buy the newspaper for one or more
days, but after having observed a drop in the asset prices would not buy it any
longer for some time after, since she expects it would still report low prices.
Another way in which dissonance may be reduced if negative news on the share-
price index appear, could be to sell the shares. However, this remedy to dissonance
is certainly more costly (selling shares would entail a transaction cost, plus the agent
should take another decision on how to invest the money realized from the sale) than
simply ignoring information. Furthermore, this behavior would also contrast with
prospect theory: people tend to hold on their shares when they are doing badly.
This behavior is a well documented empirical regularity (see, e.g., Odean, 1998).
Therefore we will explore the relation between the stock market performance and
?nancial newspaper sales in the following.
By using cointegration techniques, we ?nd that our data (we have monthly ob-
servations from 1978 to 2003) lend support to the cognitive dissonance hypothesis.
The share-price series and the series of the ?nancial newspaper sales are cointegrated
(i.e., they move together), and the former causes (in the sense of Granger, 1969) the
latter. Moreover, the relation is a positive one, that is, an increase in share prices
stimulates sales of the ?nancial newspaper.
We also analyze whether more ‘standard’ hypotheses about information acqui-
sition are consistent with the data. For instance, agents may buy newspapers to
acquire information about the stock market, in order to improve their trading in
the market, or reoptimize their portfolio (the newspaper may report detailed in-
formation about share prices, useful to better calculate own portfolio allocation).
According to this view, one should expect to ?nd that the ?nancial newspaper sales
increase with the volumes traded in the market.
Another prediction of the rational model is that the proportion of informed
2

---

individuals increases with price noise (see for example Grossman and Stiglitz, 1980).
This is due to the fact that the higher the level of noise, the less informative the
price system is, and therefore the more valuable information is to traders.
However, we ?nd that the volumes traded in the stock exchange are not coin-
tegrated with the ?nancial newspaper sales, nor is there any evidence of a causal
relation between stock market volatility and newspaper sales. Therefore our analysis
does not lend support to these two particular ‘rational’ explanations of information
acquisition in ?nancial markets.
We also perform a series of robustness checks (see section 3 for a discussion).
Firstly, we ?nd no evidence of the existence of reverse causality relations between the
variables. Further, we try to rule out some confounding factors which could possibly
drive the relationship between the stock market index and the ?nancial newspaper
sales. In particular, it may be that when the stock market performs well, people
increase their consumption of any type of newspaper. We perform a cointegration
analysis between the stock price index and the sales of sports newspapers (which are
probably the class of newspapers most distant from the ?nancial ones). We ?nd no
evidence of cointegration between these variables, which suggests that, unlike the
consumption of ?nancial papers, the consumption of sports papers does not track
stock market performance. Finally, we show results of a causality analysis between
the Italian industrial production (used as a proxy for economic activity) and Sole
24Ore sales. The two series are not cointegrated nor is there a trace of a causal
relation between them.
In order to provide additional evidence on the e?ect of interest, we replicated
our analysis using data from the United Kingdom (see section 4). Our results show
that there is evidence of a cointegration relation between the UK stock market index
and the Financial Times circulation, and that the former Granger-causes the latter.
The advantages of using Italian data are two-fold. First of all, data for newspaper
sales are available on a longer sample period (in the UK data there are several missing
observations before 1985, therefore we only use the sample period without missing
data). Secondly, the slower di?usion of the internet in Italy with respect to other
countries allows us to credibly consider the ?nancial newspaper as the main source
3

---

of information on ?nancial markets during most of our sample period.
Our paper is related to the recent and fertile literature on economics and psy-
chology. More particularly, a number of distinct models have been developed which
are able to explain biases in the acquisition of information (see Harmon-Jones and
Mills (1999) for a review of the social psychology literature and Akerlof and Dickens
(1982) for the ?rst formalization of cognitive dissonance within an economic model).
Among the more recent contributions, Rabin and Schrag (1999) explain the exis-
tence of distortions in agents’ information through cognitive mistakes, and Carrillo
and Mariotti (2000) explain ‘anomalous’ attitudes to information through strategic
decisions of agents, who choose to be ignorant in order to discipline their future
behavior.
Yet another approach explains cognitive dissonance by assuming that the agent’s
beliefs enter directly her expected-utility function. This approach has been pioneered
by Akerlof and Dickens (1982) and recent contributions include K¨oszegi (2006),
Eliaz and Spiegler (2006), Yariv (2005). In this approach, information can be used
by agents to improve their decisions, but it can also a?ect their beliefs. An agent
who maximizes a standard expected-utility function would not refuse to have free
information because this would allow her to take ‘better’ decisions, but an agent
whose beliefs enter her expected-utility function may decide to ignore information
(or to re-interpret it) so as to preserve her beliefs. ‘Anomalous’ behavior with respect
to information follows from the type of beliefs that the agent has. For instance, if the
agent has a preference for consistent beliefs, her utility increases when her beliefs are
con?rmed, and decreases when they are invalidated: the agent may want to actively
acquire information of the former type, and to ignore (or manipulate) information
which leads to the latter situation. Or, if the agent’s utility increases with the belief
the agent holds about herself, the agent may want to ignore any information which
would lead her to revise downwards the judgment of her abilities.
Although we do not venture into a theoretical model of our ?ndings, we speculate
that the latter approach might naturally lead to agents’ behavior consistent with
our empirical ?ndings. Suppose that an agent’s expected utility includes not only
the performance of the assets she holds, but also her beliefs on her abilities as
4

---

an investor. Then, our agent should be eager to acquire positive news about the
performance of her assets and would instead prefer not to see the negative news. In
other words, she would buy the ?nancial newspaper in times of high share prices
and not buy it in times of low share prices.
In this stream of the literature, Karlsson et al. (2005) present a model of belief
manipulation. When facing a changing environment, agents choose between two
psychological states: they can be either attentive and actively seek information,
or inattentive and avoid information. The authors ?nd that for some parameter
values their model gives rise to what they term ‘ostrich e?ect’ (and we call cognitive
dissonance): in ‘bad times’, individuals choose to be inattentive (and put their heads
in the sand like ostriches), while in ‘good times’ they choose to be attentive.
What makes their paper similar to ours is that they also investigate this question
empirically by looking at share prices data. They ?nd evidence that the aggregate
number of daily logins to investors’ online accounts is positively related to the stock
exchange prices, implying that investors are more likely to check the value of their
portfolio when the market is up. This ?nding therefore provides additional evidence
in favor of the ‘cognitive dissonance’ hypothesis on a di?erent dataset and with a
di?erent methodology than ours.
There are two main di?erences between Karlsson et al. (2005) and the present
paper. Firstly, and more importantly, we do not limit ourselves to investigating
whether the data support the ‘cognitive dissonance’ (or ‘ostrich behavior’) hypoth-
esis, but we also investigate competing hypotheses, according to which information
is acquired in order to improve decision-making. Secondly, we make use of di?erent
econometric methods. They have daily data for a relatively short period of time,
and they limit themselves to simply regress the aggregate number of daily logins on
the relevant share price indices; instead, we have monthly (instead of daily) obser-
vations but much longer (26 years) series of data, and we use more sophisticated
econometric techniques which also allow us to investigate the (Granger-)causality
link between the variables at issue.
Another strand of empirical literature analyzes the relationship between infor-
mation and the ?nancial market from a di?erent perspective. For instance, Funke
5

---

and Matsuda (2006) look at the impact of macroeconomic news on share prices com-
paring Germany and the US, whereas Andersson et al. (2009) look at the response
of German bonds to ECB news releases.
The remainder of the paper is as follows. Section 2 describes the data; section
3 presents our econometric methodology and results, and carries out a series of
robustness checks; section 4 considers UK data and section 5 concludes.
2
Discussion of the data
We use average monthly values from 1978 to 2003 for the Datastream price index
of the Italian stock market. This index is built by taking the ?rst 90% of all the
companies quoted on the Italian stock exchange taken in decreasing order of capi-
talization, disregarding in this way the small companies. We also consider a series of
total monthly trading volumes corresponding to the Datastream stock price index.
This series is only available from 1986.
As to the sources of information on the ?nancial market, we use data on the
sales of Il Sole 24Ore, which is by far the largest ?nancial (daily) newspaper. The
source of these data is the ADS (Accertamenti Di?usione Stampa) dataset, which
is publicly available (on paper) since 1976. ADS collects and certi?es the publish-
ers’ declarations on the number of copies sold and printed and on the number of
subscriptions. The Italian market for ?nancial newspapers is essentially composed
of three titles: Il Sole 24Ore, Italia Oggi, and MF. However, Italia Oggi and MF
have a much lower circulation compared to Il Sole 24Ore, which has historically al-
ways been the Italian ?nancial newspaper and accounts for over 90% of this market.
Therefore we only consider the sales of Il Sole 24Ore for the purpose of our analy-
sis. In particular, consistently with the stock market data, we use monthly average
sales from 1978 to 2003. Note that we use sales instead of total circulation because
the latter includes also subscriptions. We do not want to consider subscriptions to
avoid capturing the behavior of professional investors, who typically get access to Il
Sole 24Ore through annual subscriptions. (In Italy, subscriptions account for a very
small part of newspapers sales anyway.)
6

---

It could reasonably be argued that a ?nancial newspaper is not the only way
to acquire information on the stock market. Other ?nancial publications, national
newspapers, and internet services could to some extent be substitute channels of
information acquisition with respect to Il Sole 24Ore. However, there are good
reasons to think that these alternative information sources are not crucial for the
purposes of our analysis. First of all, national newspapers of general information do
not seem to be close substitutes for Il Sole 24Ore, because of their lower coverage
and level of in-depth analyses on ?nancial information.
With the advent of internet, online news have become an alternative channel to
written publications. However, since our data go back to 1978, internet was not even
existing or widespread for most of the period under consideration. The development
of internet access is a relatively recent phenomenon in Italy, and internet usage was
fairly limited even until the year 2000. The share of internet users over the total
population was around 1% in 1997, and started to increase sharply only after the year
2000, when it reached 22% of the population (source: Computer Industry Almanac).
We also performed the econometric analysis on a restricted sample where we have
deleted the last three years of observations, in order to eliminate the period where
internet started to be an alternative source of ?nancial information.
Another source of information on ?nancial markets is provided by private agen-
cies (e.g. Reuters or Bloomberg) which charge a subscription fee for their service.
Therefore they are generally targeted to professional investors who need to have con-
stant and detailed information about instantaneous variations in the stock prices.
The target readership of a ?nancial newspaper like Il Sole 24Ore is instead mainly
composed of individual investors who want to ?nd more detailed information about
the ?nancial assets they hold, obtain relevant news, and read experts’ analyses.
Since individual, non-professional investors are more likely to exhibit a ‘cognitive
dissonant’ behavior, we focus therefore on Il Sole 24Ore, which is probably the main
source of information for this type of investors (and - as said above - we do not
consider subscriptions, to focus on non-professional investors’ decisions).
In order to get a ?rst glance at the data, in Figure 1 we plot the series of sales of
Il Sole 24Ore against the series of the Datastream index. In order to interpret the
7

---

300000
4000
3500
250000
3000
200000
2500
150000
2000
1500
100000
1000
50000
500
0
0
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
2
2
2
2
9
9
9
9
9
9
9
9
9
9
9
9
9
9
9
9
9
9
9
9
9
9
0
0
0
0
7
7
8
8
8
8
8
8
8
8
8
8
9
9
9
9
9
9
9
9
9
9
0
0
0
0
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
Sole 24 Ore
Index
Figure 1: Sales of Il Sole 24Ore and stock market index (Note: the scale for Sole is
on the left axis and the scale for the index is on the right axis).
increasing trend of the stock market index, it should be taken into account that the
Italian stock exchange has grown considerably in the 1980s and even more in the
1990s. Mutual funds were introduced in 1983, but it is only from the second half of
the 1980s that they started to become a widely held ?nancial instrument. Household
participation in equity markets increased from 26.43% in 1985 to 38.19% in 1995
and to 48.24% in 1998 (source: Pelizzon and Weber, 2004, based on information
from a Bank of Italy SHIW survey).
A ?rst look at the raw data series seems to anticipate the results that we illustrate
in the next section: this graph suggests that there is some kind of relation between
the sales of Il Sole 24Ore and the stock market index. The investigation of the
existence and the direction of this relationship is the object of the next section
where we use natural logarithms of the two series. More precisely, the two variables
under consideration in the next section are the logs of Il Sole 24Ore sales (log Sole)
and the log stock market index (log index).
8

---

3
Econometric methodology and results
3.1
Main ?ndings
Our focus of interest is the relation between the Sole 24Ore sales and the stock
price index. More precisely, we would like to ?nd out whether a decline in the
stock market leads to a reduction in the Sole 24Ore sales and, hence in turn, a
rise in stock prices leads to increased newspaper sales. Such an e?ect may not be
instantaneous, of course, and in fact may not even be observed within a month which
is our observation frequency. In other words, an increase or downturn in the stock
market may only lead to a gradual adjustment in Sole 24Ore sales. In this case,
if we observe an increase in stock prices, we can predict that in the next months
Sole 24Ore sales will rise. In other words, we can improve our predictions of future
newspaper sales by taking into account the change in stock prices. This kind of
relation was formalized by Granger (1969).
Formally, a time series variable x is called Granger-causal for a variable y if the
forecasts of y can be improved by taking into account past and present information
in the x series. Suppose the two variables are generated by a vector autoregressive
process of order p (VAR(p)),
?
?
?
? ?
?
?
?
y
p
?
y
?
? t ?
11,i
?12,i
t?i
yt
= Dd
?
? ?
?
?
?
t +
+
.
xt
i=1
?21,i ?22,i
xt?i
?xt
Here dt is a vector of deterministic terms such as a constant and seasonal dummy
variables and D is the corresponding coe?cient matrix. The ?kl,i’s are the coe?-
cients of the VAR process and [?yt, ?xt] is a serially uncorrelated error process with
zero mean. Clearly, this is a reduced form model. Since forecasts are based on
reduced forms, we can analyze Granger-causality by testing restrictions on the VAR
coe?cients. For example, x is not Granger-causal for y if ?12,i = 0 for i = 1, . . . , p.
In principle, such a hypothesis is easy to test using standard ?2 or F ?tests under
usual assumptions. In a time series context the properties of these tests depend on
the trending properties of the variables, however. In particular, it matters whether
the variables have stochastic trends generated by unit roots in the autoregressive
operator and whether they are cointegrated and, hence, are driven by a common
9

---