Information rigidities, inflation perceptions, and the media: lessons from the euro cash changeover.
Lamla, Michael J. ; Lein, Sarah M.
I. INTRODUCTION
The introduction of the euro coins in January 2002 led to a
surprising surge in inflation perceptions in many euro area countries.
For example, the balanced statistic of inflation perceptions published
by the European Commission for Germany rose from about 30 in the end of
2001 to 60 at the beginning of 2002 and remained at a historically high
level for a protracted period of time. Using quantified figures
calculated from this qualitative survey, consumers' inflation
perceptions soared from 2.4% in December 2001 to 6.1% in May 2002. (1)
This jump in inflation perceptions is puzzling because at the same time
official HICP rates fell from 1.4% to 1.1%.
This mismatch between inflation perceptions and actual inflation
rates after the euro cash changeover serves as a welcome natural
experiment to test for the influence of media reporting on
consumers' opinions. This paper argues that, while the mismatch is
difficult to reconcile with fully rational consumers in a world of
perfect information, it is consistent with rationally inattentive
consumers who obtain their information from the media.
Theoretically, the idea that people have imperfect information
about economic developments can be motivated by rational inattention.
For example, Sims (2003) argues that consumers face cognitive capacity
constraints and cannot digest all available information. Consequently,
since individuals lack the capacity to absorb all publicly available
information, the coding of information, e.g., the visibility of an
article in a newspaper, becomes relevant and may induce a common
reaction in the population.
Media may influence the public opinion as people seek for a
cost-effective source to update their information sets. Indeed, watching
a TV broadcast or reading a newspaper is much less time-consuming than
searching for new information without these sources. Sims (2010)
highlights that while people do not actively search and track the latest
information regarding price developments, for instance, any headlines
covered in leading media may attract their attention and bring them to
update their information set. In addition, media outlets help to give an
interpretation to economic news and therefore do part of the job of
processing economic information. This is corroborated by survey
evidence. According to Blinder and Krueger (2004), people obtain their
information on current economic conditions mainly through the media,
especially through TV broadcasts and newspapers. More precisely, 82% say
that they receive their information from the TV and 52% from newspapers.
Using a detailed media data set, we show that exaggerated media
reporting significantly contributed to the jump in inflation
perceptions. According to our estimates, media reporting explains more
than 15% of the jump in inflation perceptions observed in January 2002.
Actual inflation accounts for only 1%. We also find that the media
reporting channel becomes particularly relevant for explaining the
dynamics of inflation perceptions after the cash changeover.
Furthermore, only articles about rising inflation fueled inflation
perceptions.
This article is related to Eife and Coombs (2007), who show that
media reporting intensified around the introduction of the euro. They
conjecture that the media shaped consumers' perceptions. In a
similar fashion, Del Giovane and Sabbatini (2006) collected media data
for Italy until 2003 and report a positive correlation between the
amount of news about inflation and the level of inflation perceptions.
This paper extends these ideas. Both papers suggest a positive
correlation. However, it is not clear why the amount of news should
increase inflation perceptions. Hence to analyze the impact of news it
is necessary not to count the amount of news but to identify its
content. We are able to identify the content using a comprehensive data
source of daily media data over 10 years for 26 media outlets. According
to our estimates, only articles about rising inflation fueled inflation
perceptions.
Understanding how inflation perceptions are formed is also useful
for gaining insights into consumers' economic behavior and the
potential feedback on other economic variables. As argued in Eife and
Meier (2007), high inflation perceptions may have real effects. They
show that the contraction in the restaurant sector experienced in the
months after the changeover is partly due to the observed overshooting
of inflation perceptions. Furthermore, higher perceived inflation rates
may lead to an underestimation of the purchasing power of households and
therefore to a reduction in spending (see Hofmann et al. 2006; Stix
2009). Inflation perception might also feed into inflation expectations,
which have consequences for wage claims, saving, and investment
decisions (Ehrmann 2011). In addition, inflation expectations may lead
to higher future inflation rates, as they may prove self-fulfilling
(Leduc, Sill, and Stark 2007). (2) Thus, although inflation perceptions
may play an important role in determining consumers' economic
actions, we still know remarkably little about the formation of
inflation perceptions (Lein and Maag 2011).
Our paper is also related to the literature that analyzes media
effects. A large body of the literature in political economy and media
science deals with the impact of media on peoples' decisions. For
example, DellaVigna and Kaplan (2007) find that media can influence
voting behavior. Similarly, Hetherington (1996) posits that media
consumption and attention through the mass media negatively shaped
voters' retrospective economic assessments in the 1992 U.S.
presidential election. Other studies find that the media tend to give a
bias to economic news in general (Gentzkow and Shapiro 2010; Groeling
and Kernell 1998; Mullainathan and Shleifer 2005; Shah et al. 1999).
Regarding the impact of media on economic figures, Carroll (2003), Lamia
and Lein (2014), and Lamia and Maag (2012) provide empirical evidence
that media reporting influences the formation of consumers'
inflation expectations and disagreement.
Furthermore, our paper is related to the literature explaining
inflation perceptions. Determinants of inflation perceptions that have
been tested in the literature are ex ante expectations, complexity of
conversion rates, asymmetry in the perception of price movements, price
movements in frequently bought items, and the proportion of people who
convert the euro price back to the old currency. We discuss these
briefly in the following paragraphs.
Consumers like to see their ex ante expectations confirmed.
Traut-Mattausch et al. (2004) present experimental evidence that links
high inflation perceptions in 2002 to the existence of a priori
expectations of high price increases before the cash changeover.
However, a priori expectations, although significant, have not been able
to explain the break in the tight relationship between perceived and
actual inflation (Doehring and Mordonu 2007).
Other explanations look at the process of converting the old
currencies into euro prices. Ehrmann (2011) compares several euro area
countries and finds that the gap is larger in countries with simpler
conversion rates, where people are more aware of price increases. Dziuda
and Mastrobuoni (2009) find that the longer people stick to converting
the euro prices into their old currency, the more likely it is that they
will overestimate current inflation.
Consumers may pay more attention to frequently bought products and
products that experience price increases. Thus the weighting that these
products are given in inflation perceptions may be disproportionately
high. There is not much evidence, however, that a strong rise in the
prices of frequently bought products can explain the gap between
perceived and actual inflation in the aftermath of the euro cash
changeover. For example, Aucremanne, Collin, and Stragier (2007) find no
evidence that strong rises in the prices of frequently bought products
can explain the rise in inflation perceptions. Doehring and Mordonu
(2007) also show that an out-of-the-pocket expenditure index does not
perform any better in explaining inflation perceptions than the
all-items HICP. Furthermore, there is evidence that consumers pay more
attention to price increases than to price reductions (Vogel, Menz, and
Fritsche 2009).
The remainder of this paper proceeds as follows: Section II
describes the data; Section III presents the econometric results of our
analysis, detailed results for different socioeconomic groups, and
robustness checks; and Section IV concludes.
II. DATA
To quantify the intensity of media coverage, we rely on data kindly
provided by the media research institute Mediatenor. The data comprise
articles and media releases on a daily frequency for the time span
January 1998 to September 2007 in Germany, covering statements dealing
with inflation that are at least five lines long in the case of printed
media, and last at least 5 seconds for television broadcasts. (3) The
coding is based on the standards of the media content analysis. (4) We
use only news in Germany about German inflation. Furthermore, the
content analysis of the reports allows us to identify whether a specific
news item states that inflation is rising or falling. (5)
In addition, we employ simple count variables that capture how
often a specific terminology is mentioned in the media. These variables
are mainly used as a test for robustness of our main results. The count
measures are obtained by searching through LexisNexis, an online
database of media articles. First, we count the articles using the term
"Teuro" (Teuro). "Teuro" is a concatenation of the
words "teuer," German word for "expensive," and the
word euro. Analogously, we count the expression "euro
introduction" (euro). The latter per se does not contain a
particular tone, as it just reminds the public of a particular event
related to their currency. The word "Teuro," however, clearly
presumes that inflation has been and/or will be rising and is related to
the cash changeover in 2002. Given that there is no evidence that the
introduction of the euro affected prices in Germany significantly, the
Teuro discussion serves as an example of an exaggeration made by the
media. (6)
As the measure for perceived inflation we employ survey data
collected by the EU Consumer Survey. Inflation perceptions are captured
by asking households: "How do you think that consumer prices have
developed over the last 12 months? They have...." Respondents
express their beliefs on a five-option ordinal scale: "risen a lot,
risen moderately, risen slightly, stayed about the same, fallen."
We use the balance figures as calculated by Eurostat, i.e., the
difference between positive and negative answers (in percentage points
of total answers). (7) Given these data sources we can directly map news
from Germany on inflation in Germany with inflation perceptions of
German consumers.
[FIGURE 1 OMITTED]
The series of inflation perceptions tracks the HICP inflation rate
relatively closely prior to 2002 (Figure 1, left panel). After the euro
cash changeover, this co-movement between the two series breaks down.
Regarding media coverage, the link between inflation and media reporting
is not clear either. Even though a high volume of media reporting
usually coincides with a high level of inflation, the reverse is usually
not true. There can be high media coverage and low rates of inflation
and vice versa. Examples for this phenomenon can be found in mid-2002 as
well as at the beginning of 2003 (right panel). Thus, media coverage
does not necessarily co-move with inflation.
To gain more insights into these relationships, we disentangle all
reports into those dealing with rising inflation news (Rising Inflation)
and falling inflation news (Falling Inflation) and plot them together
with HICP in the left panel in Figure 2. We can observe that if
inflation is rising, then the media reports that inflation is rising,
and vice versa. Thus, media agencies capture the overall dynamics
rightly. However, the amount of reporting does not necessarily match the
magnitude of price changes. Comparing the spikes in 2002 and 2004 shows
that although inflation was as high, the coverage in the media was very
different. Moreover, it seems that there is a greater propensity to
report on rising inflation than on falling inflation.
The sharp rise in inflation perception corresponds with the
repeated use of "Teuro." The right panel in Figure 2 shows the
relationship between Teuro and inflation perceptions. Given the fact
that the word "Teuro" was chosen "the word of the year
2002," it is not surprising that the euro introduction was one of
the main topics in these year reviews. During the year 2003, both media
reports and perceived inflation fell back to a relatively low level.
III. ESTIMATION AND RESULTS
This section comprises two parts. First, we quantify the impact of
media on inflation perceptions. Second, we test whether the impact of
media reporting has changed over time.
[FIGURE 2 OMITTED]
A. Linear Framework
Our starting regression rests on the empirical model proposed by
Doehring and Mordonu (2007). The determinants are lags of perceptions,
inflation expectations, HICP inflation, and a dummy variable controlling
for the euro cash changeover. We estimate models in the following form:
[[pi].sup.perc.sub.t] = [alpha] +
[[beta].sub.1][[pi].sup.perc.sub.t-1] +
[[beta].sub.2][[pi].sup.exp.sub.t-6] + [[beta].sub.3][[pi].sub.t] +
[[beta].sub.4][D.sub.>2002] + [GAMMA] [Media.sub.t] +
[[epsilon].sub.t].
Inflation perceptions are denoted by [[pi].sup.perc] and inflation
expectations by [[pi].sup.exp]. We use a 6-month lag of expectations as
an explanatory variable because experimental evidence has shown that
expectations about price developments influence how inflation is
perceived (Traut-Mattausch et al. 2004). (8) Current inflation is
included as the growth rate of the HICP ([pi]). (9) The dummy variable
([D.sub.>2002]) is constructed according to Doehring and Mordonu
(2007). It is 0 until 2002 and 1 afterwards. Media represents a vector
of media variables that will be added to the specification.
The Teuro debate in the media had a significant impact on inflation
perceptions. This is shown in the first column in Table I. The variable
counting the number of times "Teuro" was mentioned in the
print media has a significant positive impact on inflation perceptions.
This is reasonable, as the main message of those articles was indeed to
"warn" the public of rising prices with respect to the
introduction of the euro. It is notable that the discussion on the
introduction of the euro itself reveals no such impact. The results for
the control variables are in line with the findings in Doehring and
Mordonu (2007). Finally, people also incorporate actual statistical
information as proxied by the HICR
News on rising inflation significantly raises inflation
perceptions, while news on falling inflation does not have a
statistically measurable impact (column (2)). There is thus an asymmetry
in the relationship between media reporting and inflation perceptions.
Interestingly, the inflation figure is no longer significant. This
therefore suggests that the statistical information is provided by media
reports and that consumers do not react to statistical news not
discussed in the media. (10) Given the evidence on macroeconomic
illiteracy among consumers (Blanchflower and Kelly 2008), this result is
not surprising; consumers have difficulties in interpreting an inflation
figure as such and hence need the media reports, which help them to
understand what the figure means and what it implies--for example, their
consumption expenditures.
In our next step, we split the sample into two subsamples: one
before the euro cash changeover and one afterward. Interestingly, the
media had no explanatory power before the cash changeover, but the HICP
figure was statistically significant (column (3)). By contrast, after
the introduction of the euro, consumers relied on information provided
by the media as well as on their past expectations (column (4)). This
suggests that the effect of reporting about rising inflation on
inflation perceptions is a phenomenon that arose after the cash
changeover. (11) Moreover, the result that inflation expectations are
insignificant in the period before the cash changeover suggests that
there is no evidence for the expectation confirmation bias before 2002
(Traut-Mattausch et al. 2004). In sum, the estimation result clearly
indicates that the media are a statistically important driver of
inflation perceptions. News on rising inflation has an especially strong
influence.
One could contend that media reports about inflation may influence
inflation perceptions and that those perceptions might also drive media
reporting or, put differently, media agencies might cater to the
prejudices of their readers and therefore react to inflation
perceptions. We address this endogeneity problem in various ways. First,
we use an instrumental variable approach. Second, we perform Granger
causality tests. Both suggest that there is a causality from media to
perceptions and that our results are not driven by reverse causality.
We apply two-stage least squares (2SLS) and general methods of
moments (GMM). The variables rising inflation and falling inflation are
instrumented by their own four lags and the remaining exogenous
variables in the model (columns (5) and (6)). For the 2SLS the [chi
square]-test of overidentification cannot reject the null hypothesis
that our instruments are valid. The test statistic is 5.14, the
associated p value is .53. For the GMM estimation the corresponding
I-test has a p value of .63 and leads to the same conclusion. The
relevance of instruments is supported by first-stage Shea's (1997)
partial [R.sup.2] statistics, which amount to .25 for rising inflation
and to .2 for falling inflation. Concerning our estimates, the size of
the effect of news about rising inflation becomes even larger when using
the instrumental variables approach.
Granger causality tests do not indicate that the results are driven
by reverse causality. As a further robustness check, we estimate the
impact of media reporting in a dynamic setting. A vector autoregression
(VAR) setup allows us to account for the dynamics between the different
variables, especially between perceived inflation and the media
variables. Hence, these variables are set to be endogenous in the
system. Exogenous variables are the 6-month lag of expectations, the
HICR and the changeover dummy. We use four lags according to the Akaike
information criterion. The results of the Granger causality analysis are
presented in Table 2. Lagged media variables significantly affect
perceptions but the reverse causality link is not statistically
significant. This confirms that, although reverse causality might be
present, the dominating channel is the influence of media on inflation
perceptions and there is no measurable feedback effect from the level of
perceptions.
Our results have shown that media reports are statistically
significant and robustly related to inflation perceptions. In the
following, we show that they are also economically important. Media
reporting explains more than 15% of the change in inflation perceptions
observed during the cash changeover in January 2002. Actual inflation
accounts for only 1%. These estimates are obtained by fitting the
observed inflation perceptions in December 2001 and January 2002 to the
equation estimated in column (2) of Table 1 and taking first
differences.
Furthermore, media reporting on rising inflation is, after the lag
of perceptions, the variable with the highest impact on inflation
perceptions. To quantify the relative economic importance of the media,
we report the impact of each variable on inflation perceptions based on
the impulse of a shock of one unit standard deviation of the respective
series. Figure 3 illustrates this. As expected, the autoregressive
parameter is the highest on impact. Above that, the figure offers
another interesting insight: the response to the rising inflation
variable is found to be much higher compared to the remaining
explanatory variables. To sum up, the estimation results give wide
support for the effect of the media on the perceptions of consumers.
B. Socioeconomic Characteristics
In the next section, we test whether the results established so far
hold for different socioeconomic groups. Several studies provide
empirical evidence that socioeconomic characteristics matter.
For example, Palmqvist and Stroemberg (2004) find a u-shaped
relationship between inflation perception and age as well as income in
Sweden. People's inflation perceptions become more accurate into
middle age and then deteriorate when they become elderly. They find
similar results for inflation expectations. Stix (2009) and Dziuda and
Mastrobuoni (2009) argue that household income, education level, or age
are factors determining changeover-induced inflation perceptions.
Malgarini (2009) investigates the relationship between personal
characteristics and the overestimation of inflation, showing that
socioeconomic characteristics matter for the degree of overestimation of
inflation figures. He notes that the degree of overestimation becomes
lower as the level of education rises. Furthermore, he shows that more
optimistic respondents are prone to a lower degree of overestimation.
Blanchflower and Kelly (2008) report very high nonresponse rates to
inflation perceptions in surveys among the least educated, females, the
poorest and younger individuals. Furthermore, groups with biased
perceptions form biased expectations as well.
As Blinder and Krueger (2004) show in a survey for the United
States, people receive the bulk of information from media usage (mainly
TV; followed at a large distance by newspapers), but they do not
actively search for information on economic issues. People with higher
incomes as well as a higher education are in general better informed.
Furthermore, ideology plays a large role in the formation of public
beliefs. (12) Therefore, the different usage structure of media by
different household types might play an important role in explaining
differences in perceptions.
[FIGURE 3 OMITTED]
The regressions summarized in Table 3 represent the main
regressions estimated in Table 1, the difference being that the
dependent variable measures inflation perceptions, calculated separately
for each socioeconomic group. The explanatory variables remain the same.
We focus on gender, earnings, education, and age. (13)
In general, all socioeconomic groups react to media reports.
Interestingly, there is not a single group that would not react at all.
However, there are some differences across socioeconomic groups. For
more educated people, the results show a stronger link between HICP and
inflation perceptions. At the same time, they react to news on the
"Teuro" as well as to news on rising inflation. On the one
hand, this implies that educated people rely more on existing
statistical figures. Furthermore, consumers with further education
(above secondary), do not respond significantly to news about rising or
falling inflation.
With respect to gender, the estimated coefficients are very
similar, independent of the regression setup. For different age groups,
we find some evidence for an inverted u-shape relationship, in line with
Palmqvist and Stroemberg (2004). For instance, consumers in the age
category ag2 = 30-49 neither respond to news about rising or falling
inflation in the whole sample period, nor do they respond in the period
after 2002 to news about rising inflation where all other groups
responded. However, they seem to follow the published inflation figures
closely. This might be due to the fact that these groups are less
affected by biased media reporting, because, for example, they choose
media outlets that report in a less biased manner.
Consumers in the third income quartile respond most to news on
inflation as well as HICP for the full horizon. However, in the
aftermath of the cash changeover income groups re2 and re4 respond
substantially stronger.
While those results may in part be less clear cut, they certainly
suggest some differences in the response due to socioeconomic
characteristics. The better educated people are, the more they rely on
real HICP figures and are less influenced by biased media reporting.
C. Nonlinear Framework
In this section, it is shown that the relationship between media
reporting and perceptions differs during the euro cash changeover
period, without ex ante imposing a break in the relationship. This
section provides additional evidence on the structural break examined
before, and serves as a last robustness check.
To do this, we use an alternative way of testing the effect of the
introduction of the euro that allows us to estimate its effect more
flexibly. We use a logistic smooth transition regression (LSTR), where
we first test for nonlinearity in the relationship between media
reporting and inflation perceptions and second, estimate a model that
defines the coefficient of media reporting as a function of various
indicators, e.g., the time trend.
The LSTR model is defined as follows:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (1)
where [gamma] > 0 is an identifying restriction. A more detailed
explanation of the model is presented in Terasvirta (2004). We test for
the LSTR1 (K = 1) and LSTR2 (K = 2). In the LSTR1 model, the parameters
change monotonically as a function of [s.sub.t] from [phi] to [phi] +
[theta]. In the LSTR2 model, they change symmetrically around the
midpoint 1/2([c.sub.1] + [c.sub.2]), where the logistic function has its
minimum. This minimum value is between 0 (for [gamma] [right arrow]
[infinity]) and 0.5 (for [c.sub.1] = [c.sub.2]). When [gamma] = 0, the
LSTR2 model is equivalent to the linear model. This implies that in the
LSTR1 model, the coefficient of interest changes from one regime to the
other, whereas in the LSTR2 model, the regime is the same for high and
low values of the transition variable [s.sub.t], but differs in the
middle range. We shall explain this in more detail below.
The transition variable [s.sub.t] can be chosen as a stochastic
variable contained in [z.sub.t], but it may also be a variable that is
not part of the set of other explanatory variables. Another interesting
case for our purposes is that [s.sub.t] is a linear trend ([s.sub.t] =
t), which yields a linear model with deterministically changing
parameters. In the following, we start our analysis by testing whether
we need to specify a nonlinear model, i.e., we test the linear model
against the LSTR model with different types (i.e., K = 1 and K = 2) and
a set of potential transition variables [s.sub.t].
The transition variable plays an important role in the economic
interpretation of our model. This variable determines the current slope
coefficient. In other words, if the model selection tests show that the
appropriate model is the LSTR1 model with media reporting, then the
relationship between our dependent variable and the explanatory
variables (in the nonlinear part of the model) changes with the
intensity of media reporting. The set of potential transition variables
should be chosen from theory. As suggested earlier, we expect that the
relationship between media reporting and inflation perceptions may
change due to more media reporting, due to the euro cash changeover, or
that during high inflation periods, the impact on perceptions was higher
than during low inflation periods. Thus, we test the linear model
against the LSTR1 and LSTR2 model, with different transition variables:
the time trend, the intensity of media reporting, and the level of
inflation.
Our model can be written as follows:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (2)
where we subsequently include the variables timetrend,
[media.sub.t-1], [[pi].sub.t], and [[pi].sup.perc.sub.t-1] into
[s.sub.t], and test the LSTR1 and LSTR2 against the linear model. The
testing procedure has two steps. First, the linear model is tested
against the nonlinear model.
Like many nonlinear models, the LSTR model is only identified under
the alternative, not the null hypothesis due to the nuisance parameters
[theta] and c. Therefore, the transition function in the LSTR model is
substituted by a third-order Taylor approximation around the null. The
model LSTR1 is assumed, which allows testing both for LSTR1 and LSTR2
(14). The test is based on the following regression equation
[y.sub.t] = [[beta]'.sub.0][z.sub.t] + [3.summation over
(j=1)] [[beta]'.sub.j] [[??].sub.t][s.sup.j.sub.t] +
[u.sup.*.sub.t]
where [z.sub.t] = (1, [[??].sub.t])' and [u.sup.*.sub.t] =
[u.sub.t] + R ([gamma], c, [s.sub.t]) [theta]'[z.sub.t] with the
remainder R([gamma], c, [s.sub.t]). The null hypothesis of linearity is
[[beta].sub.1] = [[beta].sub.2] = [[beta].sub.3] = 0, because under
[H.sub.0], the asymptotic distribution theory is not affected if an
LM-type test is used. (15) If linearity was not rejected in the first
step, the modeling cycle ends and the linear model is chosen. When
linearity is rejected, tests have to be performed to decide between the
LSTR1 and LSTR2 model. The test makes use of the auxiliary regression
described by Equation (3). The following sequence of null hypotheses is
defined:
[H.sub.4] : [[beta].sub.3] = 0
[H.sub.3] : [[beta].sub.2] = 0|[[beta].sub.3] = 0
[H.sub.2] : [[beta].sub.1] = 0|[[beta].sub.3] = [[beta].sub.2] = 0
Granger and Terasvirta (1993, Ch. 7) show that in this test
sequence, if the rejection of [H.sub.3] is the strongest of the three
tests, then a LSTR2 model is chosen. Otherwise a LSTR1 model is
selected. All three hypotheses can be rejected simultaneously, hence the
strongest rejection indicates the best choice of the model. Test results
are reported in Table 4. The first column (F) reports the p value for
the F-statistic of the first step test (linearity), the following three
columns (F4, F3, F2) report the p values of the tests for the
appropriate model, according to the null hypotheses defined in the
second-step testing sequence. The final column lists the suggested
model. Each row reports the testing results for the respective
transition variable.
The linearity tests suggested by Luukonen, Saikkonen, and
Terasvirta (1988) and Terasvirta (1994) favor the LSTR2 model with the
time trend as transition variable. The coefficient on inflation is set
to remain linear, as we expect that the impact of the true inflation
rate remains stable, when controlling for lagged perceptions and media
reporting. Furthermore, we included the inflation rate in the nonlinear
part and it turned out to be insignificant. Thus, the autoregressive
term and the intensity of media reporting are included in the nonlinear
part of the LSTR2 model.
Our results are reported in Table 5. The estimated transition
function is illustrated in Figure 4.
We find that media reporting was of minor importance before 2002;
the transition function is close to I up to 2002, which implies that the
impact of media reports is [[phi].sub.3] + [[theta].sub.3] = 0.53. Then,
during the year 2002, the transition function declines and attains its
minimum value (0.5), which implies that the impact of media reporting
rises to a maximum of [[phi].sub.3] + [[theta].sub.3] x 0.5 = 5.3. At
the same time, the persistence of perceived inflation falls from 0.67 to
-1.01 during the changeover period in 2002. Furthermore, the constant is
higher during the changeover period. These findings suggest that, during
the changeover in 2002, the dynamics of perceived inflation rates
changed, and that media reporting gained in importance. This result from
the nonlinear estimation complements the outcome from the linear
regression model which indicates a substantial break in the inflation
perception relationship in the aftermath of the cash changeover.
To go one step further, we investigate the role of the content of
media reports. If the tone of media reports triggered the change in the
relationship between inflation perceptions and media reporting, we
should find that in our data. Thus we test for the nonlinearity in our
model, adding rising inflation and falling inflation to it. Again, we
test the LSTR1 and LSTR2 against the linear model with all possible
transition variables. Our results are reported in Table 6.
Interestingly, when adding the tone variables, the nonlinearity vanishes
and the tests favor the linear model for all possible transition
variables. This suggests that the information that was responsible for
the nonlinearity has now been included and that the change in the
dynamics of inflation perceptions is indeed driven by the tone of media
reports.
IV. CONCLUSION
Consumers do not constantly track the latest statistics, nor do
they have their own macroeconomic models that they continuously update,
because there are costs involved in acquiring and processing
information. The media provide a low-cost source of information that
consumers use to update their economic information. Thus the tone in
media reports about the state of the economy feeds through into
consumers' economic perceptions. We test for the effect of media
using the example of inflation perceptions.
[FIGURE 4 OMITTED]
Employing a detailed data set on media coverage for Germany over
more than ten years, we are able to confirm that media reporting has a
strong influence on the inflation perceptions of consumers and that the
large jump in inflation perceptions after the introduction of the euro
can to a large extent be explained by the presence of news on rising
inflation.
This paper has implications for understanding the dynamics of
inflation perceptions. First, it highlights the fact that the media
affects inflation perceptions and thereby supports theoretical concepts
regarding information rigidities. This influence varies over time and
according to socioeconomic characteristics. Both issues are welcome
avenues for future research. Given the strong deviation of inflation
perceptions from official data on inflation rates in the aftermath of
the euro cash changeover, our paper also offers a policy implication: as
discussed in Eife and Coombs (2007) and Eife (2006), the authorities
should engage in active countermeasures if the perception of the public
diverges from actual inflation rates.
ABBREVIATIONS
2SLS: Two-Stage Least Squares
GMM: General Methods of Moments
HICP: Harmonized Index of Consumer Prices
LSTR: Logistic Smooth Transition Regression
VAR: Vector Autoregression
doi: 10.1111/ecin.12121
REFERENCES
Aucremanne, L., M. Collin, and T. Stragier. "Assessing the Gap
Between Observed and Perceived Inflation in the Euro Area: Is the
Credibility of the HICP at Stake?" Working Papers Series 122,
National Bank of Belgium, 2007.
Blanchflower, D. G., and R. Kelly. "Macroeconomic Literacy,
Numeracy and the Implications for Monetary Policy." Speech, Bank of
England, 2008.
Blinder, A. S., and A. B. Krueger. "What Does the Public Know
About Economic Policy, and How Does It Know It?" NBER Working
Papers 10787, National Bureau of Economic Research, Inc., 2004.
Carroll, C. D. "Macroeconomic Expectations of Households and
Professional Forecasters." Quarterly Journal of Economics, 118(1),
2003, 269-98.
Curtin, R. "What U.S. Consumers Know About Economic
Conditions." Mimeo, 2007.
Del Giovane, P., and R. Sabbatini. "Perceived and Measured
Inflation After the Launch of the Euro: Explaining the Gap in
Italy." Giomale degli Economisti, 65, 2006, 155-92.
Della Vigna, S., and E. Kaplan. "The Fox News Effect: Media
Bias and Voting." Quarterly Journal of Economics, 122(3), 2007,
1187-234.
Doehring, B., and A. Mordonu. "What Drives Inflation
Perceptions? A Dynamic Panel Data Analysis." Economic Papers 284,
European Commission, 2007.
Dziuda, W., and G. Mastrobuoni. "The Euro Changeover and Its
Effects on Price Transparency and Inflation." Journal of Money,
Credit and Banking, 41(1), 2009, 101-29.
Ehrmann, M. "Rational Inattention, Inflation Developments and
Perceptions After the Euro Cash Changeover." German Economic
Review, 12(1), 2011, 33-58.
Eife, T. A. "Price Setting Behaviour and Price Setting
Regulations at the Euro Changeover." Bank of Estonia Working Papers
2006-06, Bank of Estonia, 2006.
Eife, T. A., and W. T. Coombs. "Coping with People's
Inflation Perceptions During a Currency Changeover." Working Papers
0458, Department of Economics, University of Heidelberg, 2007.
Eife, T. A., and S. Meier. "Costly Inflation
Misperceptions." Working Papers 0455, Department of Economics,
University of Heidelberg, 2007.
Fluch, M., and H. Stix. "Perceived Inflation in
Austria--Extent, Explanations, Effects." Monetary Policy & the
Economy, (3), 2005, 22-47.
Gentzkow, M., and J. M. Shapiro. "What Drives Media Slant?
Evidence from U.S. Daily Newspapers." Econometrica, 78(1), 2010,
35-71.
Granger, C., and T. Terasvirta Modelling Non-Linear Economic
Relationships. Oxford: Oxford University Press, 1993.
Groeling, T., and S. Kernell. "Is Network News Coverage of the
President Biased?" Journal of Politics, 60, 1998, 1063-87.
Hetherington, M. J. "The Media's Role in Forming
Voters' National Economic Evaluations in 1992." American
Journal of Political Science, 40, 1996, 372-95.
Hofmann, E., B. Kamleitner, E. Kirchler, and S. Schulz-Hardt.
"Kaufkraftschwund nach der Waehrungsumstellung: Zur
erwartungsgeleiteten Wahrnehmung des T(Euro)."
Wirtschaftspsychologie, 1, 2006, 98-108.
Lamia, M. J., and S. M. Lein. Forthcoming. "The Role of Media
for Consumers' Inflation Expectation Formation." Journal of
Economic Behavior and Organization, 2014, forthcoming.
Lamia, M. J., and T. Maag. "The Role of Media for Inflation
Forecast Disagreement of Households and Professional Forecasters,"
Journal of Money, Credit and Banking, 44(7), 2012, 1325-50.
Leduc, S., K. Sill, and T. Stark. "Self-fulfilling
Expectations and the Inflation of the 1970s: Evidence from the
Livingston Survey." Journal of Monetary Economics, 54(2), 2007,
433-59.
Lein, S. M., and T. Maag. "The Formation of Inflation
Perceptions: Some Empirical Facts for European Countries." Scottish
Journal of Political Economy, 58(2), 2011, 155-88.
Luukonen, R., P. Saikkonen, and T. Terasvirta. "Testing
Linearity Against Smooth Transition Autoregressive Models."
Biometrika, 75, 1988, 491-9.
Malgarini, M. "Quantitative Inflation Perceptions and
Expectations of Italian Consumers." Giomale degli Economisti,
68(1), 2009, 53-80.
Mullainathan, S., and A. Shleifer. "The Market for News."
American Economic Review, 95(4), 2005, 1031-53.
Palmqvist, S., and L. Stroemberg. "Households' Inflation
Opinions--A Tale of Two Surveys." Economic Review, 4, 2004,
Sveriges Riksbank.
Shah, D. V., M. D. Watts, D. Domke, D. Fan, and M. Fibison.
"News Coverage, Economic Cues, and the Public's Presidential
Performance, 1984-1996." Journal of Politics, 61, 1999, 914-43.
Shea, J. "Instrument Relevance in Multivariate Linear Models:
A Simple Measure." Review of Economics and Statistics, 79, 1997,
348-52.
Sims, C. "Implications of Rational Inattention." Journal
of Monetary Economics, 50(3), 2003, 665-90.
Sims, C. A. "Rational Inattention and Monetary
Economics," in Handbook of Monetary Economics, Vol. 3, edited by B.
M. Friedman and M. Woodford. Stanford: Elsevier, 2010, 155-81.
Stix, H. "Perceived Inflation and the Euro: Evidence from an
Austrian Survey." European Journal of Political Economy, 25(4),
2009, 547-61.
Terasvirta, T. "Specification, Estimation and Evaluation of
Smooth Transition Autoregressive Models." Journal of the American
Statistical Association, 89, 1994, 208-18.
--. "Smooth Transition Regression Modelling," in Applied
Time Series Econometrics. Cambridge: Cambridge University Press, 2004.
Traut-Mattausch, E., S. Schulz-Hardt, T. Greitemayer, and D. Frey.
"Expectancy Confirmation in Spite of Discontinuing Evidence: The
Case of Price Increases Due to the Introduction of the Euro."
European Journal of Social Psychology, 34, 2004, 739-60.
Vogel, L., J.-O. Menz, and U. Fritsche. "Prospect Theory and
Inflation Perceptions--An Empirical Assessment." Macroeconomics and
Finance Series 200903, Department Wirtschaft und Politik, Hamburg
University, 2009.
(1.) Quantified inflation perceptions are kindly provided by the
Bundesbank. The calculation is described in detail in the Bundesbank
Monthly Bulletin, November 2007.
(2.) Evidence for a possible spillover from inflation perception to
inflation expectations is provided by Fluch and Stix (2005). This
finding was recently supported by the survey of the Bank of England and
Blanchflower and Kelly (2008), who conclude that consumers' price
expectations are influenced by past experience of inflation.
(3.) The following news sources are included: daily press:
Frankfurter Allgemeine Zeitung, Welt, Suddeutsche Zeitung, Frankfurter
Rundschau, Tageszeitung, Bild, Neue Zurcher Zeitung, Berliner,
Volksstimmer, Sachsische, Westdeutsche Allgemeine Zeitung, Kolner
Stadt-Anzeiger, Rheinischer Merkur; daily TV-news: ARD Tagesschau,
Tagesthemen, ZDF Heute, Heute Journal, RTL Aktuell, SAT. 1 18:30,
ProSieben Nachrichten; weekly press: Spiegel, Focus, Die Woche,
Wochenpost, Welt am Sonntag, Bild am Sonntag, Die Zeit.
(4.) See www.mediatenor.de for details on media content analysis.
(5.) This content analysis has advantages over algorithm-based
coding, because human coders automatically identify the meaning of
"rising inflation" in any textual arrangement and hence
identify, for instance, negations.
(6.) Arguably, some of the Teuro-related articles will be captured
in the variable counting news items that state that inflation is rising.
Thus, these two variables will overlap to some extent. However, some of
the articles that mention the teuro could in fact say that the euro
introduction was not inflationary and therefore the euro is not a
"Teuro." Because we cannot make sure what exactly the
Teuro-news say, we put more emphasis on the data from Mediatenor.
(7.) As a robustness check, we have quantified inflation
perceptions using the method in Lein and Maag (2011). The corresponding
econometric analysis leads to qualitatively the same results and is
available on request.
(8.) A 12-month lag produces similar results.
(9.) We also employed the out-of-the-pocket expenses index, as
calculated by Eurostat as a measure for inflation. Our motivation is
that the index should better reflect that perceptions could be more
affected by prices of frequently purchased items. The out-of-the-pocket
expenses index does not outperform the HICP in statistical terms nor
economically.
(10.) This result is not influenced by multicollinearity among the
regressors, as the correlation between the regressors is well below .6.
(11.) The effect may be persistent, as suggested by
"agenda-setting approaches" which imply a threshold effect;
once the reporting on a certain topic achieves a certain intensity, it
is perceived as a "hot topic" and remains visible for a longer
time (Vogel, Menz, and Fritsche 2009).
(12.) See also Curtin (2007).
(13.) Note that for reasons of readability we refrain from
reporting the results for the lagged dependent variable, inflation
expectations, and the constant.
(14.) This has been suggested by Terasvirta (1994, 2004).
(15.) The test statistic has an asymptotic [chi
square]-distribution when the null is valid. However, the statistic can
be distorted in a small sample, and hence an F-statistic is computed
instead, as the F-statistic has better small sample properties (the
empirical size of the test remains close to the nominal size while the
power is good), as shown by Granger and Terasvirta (1993, Ch. 7) and
Terasvirta (1994, 2004).
MICHAEL J. LAMLA and SARAH M. LEIN *
* We thank the editor and two anonymous reviewers for their
constructive comments. Furthermore, we thank Lena Drager, Ulrich
Fritsche, Matthias Vollbracht, the participants of the Swiss Society of
Economics and Statistics Meeting 2009, Geneva, Switzerland and the Royal
Economic Society Annual Meeting 2010, Guildford, UK for helpful comments
and suggestions. Financial support by the Swiss National Science
Foundation (SNSF) and the DG ECFIN is gratefully acknowledged. The views
expressed in this paper reflect solely the views of the authors and not
necessarily those of the Swiss National Bank.
Lamia: Professor, University of Essex, Wivenhoe Park, CO4 3SQ, UK;
KOF, ETFI Zurich, CH-8092 Zurich, Switzerland. Phone +44 (0) 1206
874523, Fax +44 (0) 1206 873429, E-mail mlamla@essex.ac.uk
Lein: Senior Economist, Economic Analysis, Swiss National Bank,
Borsenstrasse 15, CH-8022 Zurich, Switzerland. Phone +41 44 6313199, Fax
+41 446313901, E-mail sarah.lein@snb.ch
TABLE 1
Determinants of Inflation Perceptions
(1) (2) (3) < 2002
[[pi].sup.perc 0.891 *** 0.925 *** 0.775 ***
.sub.t-1] (0.032) (0.025) (0.069)
[[pi].sup.exp 0.161 *** 0.162 *** 0.002
.sub.t-6] (0.050) (0.039) (0.082)
[[pi].sub.t] 2.494 *** 0.547 4.969 ***
(0.716) (0.570) (1.547)
Teuro 0.036 ***
(0.013)
Euro 0.002
(0.025)
Rising 0.169" * 0.058
inflation (0.082) (0.108)
Falling -0.054 -0.083
inflation (0.081) (0.100)
[D.sub.>2002] 1.646 2.888 **
(1.575) (1.104)
Constant -5.293 *** -4.770 *** -0.725
(1.738) (1.423) (2.403)
Observations 102 102 42
(4) (5) IV (6) IV
>2002 2SLS GMM
[[pi].sup.perc 0.939 *** 0.920 *** 0.923 ***
.sub.t-1] (0.036) (0.026) (0.024)
[[pi].sup.exp 0.166 *** 0.141 *** 0.148 ***
.sub.t-6] (0.038) (0.052) (0.048)
[[pi].sub.t] 1.005 -0.277 -0.048
(0.983) (0.720) (0.681)
Teuro
Euro
Rising 0.190 *** 0.308 ** 0.274 **
inflation (0.068) (0.147) (0.127)
Falling 0.078 0.001 -0.043
inflation (0.131) (0.225) (0.213)
[D.sub.>2002] 3.460 *** 3.293 ***
(1.092) (1.038)
Constant -3.854 -4.608 *** -4.599 ***
(2.409) (1.571) (1.488)
Observations 60 102 102
Notes: Newey-West corrected standard errors in parentheses.
Dependent variable is the inflation perception at time t
[[pi].sub.perc.sub.t] [[pi].sup.exp] denotes inflation
expectations, [[pi].sup.t], the inflation rate in period t.
Media variables: teuro is the monthly number of articles
dealing with the word Teuro, euro is the number of articles
with the word euro, rising (falling) inflation is the number
of articles reporting on rising (falling) inflation.
[D.sub.>2002] is a dummy that takes the value one after
December 2001, and zero otherwise. Columns (1) and (2)
report the results for the whole sample period, columns (3)
and (4) for the sub-sample before and after 2002,
respectively. Columns (1) to (4) are estimates with OLS,
while columns (5) and (6) are instrumental variables
estimations using 2SLS in column (5) and GMM in column (6).
*** p < .01, ** p < .05, * p < .1.
TABLE 2
Granger Causality
Equation Excluded F df dfr Prob > F
[[pi].sup.perc] Rising 2.513 4 86 0.047
inflation
[[pi].sup.perc] Falling 0.972 4 86 0.426
inflation
[[pi].sup.perc] All 2.213 8 86 0.033
Rising inflation [[pi].sup.perc] 1.528 4 86 0.201
Rising inflation Falling inflation 0.971 4 86 0.427
Rising inflation All 1.174 8 86 0.323
Falling inflation [[pi].sup.perc] 0.893 4 86 0.471
Falling inflation Rising inflation 2.594 4 86 0.041
Falling inflation All 1.684 8 86 0.113
TABLE 3
Media and Socioeconomic Characteristics
Variable (1) (2) (3) (4) (5)
Tot re1 re2 re3 re4
[pi] 1.826 ** 1.898 ** 1.763 * 2.031 ** 1.718 *
(0.686) (0.906) (0.946) (1.005) (0.902)
[D.sub. 0.935 1.009 1.501 1.536 1.562
>2002] (1.153) (1.316) (1.471) (1.595) (1.451)
[pi] 2.494 *** 2.689 *** 2.950 *** 3.282 *** 2.315 *
(0.716) (0.974) (1.050) (1.064) (0.993)
Teuro 0.036 ** 0.042 ** 0.075 * 0.067 ** 0.039 *
(0.013) (0.021) (0.030) (0.021) (0.020)
Euro 0.002 -0.007 0.023 0.026 0.013
(0.025) (0.023) (0.031) (0.035) (0.038)
[D.sub. 1.646 1.962 1.133 2.290 2.497
>2002] (1.575) (1.508) (2.112) (1.870) (1.896)
[pi] 0.547 0.935 0.195 0.707 0.216
(0.570) (0.770) (1.092) (0.994) (0.883)
Rising 0.169 ** 0.154 0.202 * 0.213 ** 0.184 *
inflation (0.082) (0.095) (0.103) (0.097) (0.108)
Falling -0.054 0.002 -0.053 0.003 -0.076
inflation (0.081) (0.101) (0.127) (0.115) (0.118)
[D.sub. 2.888 * 3.069 ** 3.138 * 3.868 * 3.527 *
>2002] (1.104) (1.245) (1.625) (1.497) (1.406)
<2002 4.969 *** 6.499 ** 4.142 5.517 * 7.065 ***
[pi] (1.547) (1.640) (2.501) (2.592) (1.992)
Rising 0.058 0.068 0.040 0.106 0.029
inflation (0.108) (0.130) (0.153) (0.131) (0.118)
Falling -0.083 -0.013 -0.171 -0.159 -0.131
inflation (0.100) (0.120) (0.188) (0.135) (0.145)
[greater 1.005 0.719 -0.281 1.784 0.001
than or (0.983) (1.537) (1.849) (1.681) (1.642)
equal to]
2002
[pi]
Rising 0.190 *** 0.136 * 0.273 *** 0.209 * 0.261 *
inflation (0.068) (0.080) (0.094) (0.112) (0.110)
Falling 0.078 0.144 0.196 0.227 -0.003
inflation (0.131) (0.202) (0.189) (0.189) (0.180)
Variable (6) (7) (8) (9) (10)
ed1 ed2 ed3 ag1 ag2
[pi] 1.516 * 2.039 * 2.778 * 1.425 * 2.085 *
(0.648) (0.789) (1.382) (0.639) (0.850)
[D.sub. 0.926 1.341 0.494 1.471 1.095
>2002] (1.255) (1.250) (1.859) (1.345) (1.325)
[pi] 2.158 ** 2.853 ** 3.724 * 2.012 ** 2.804 **
(0.689) (0.826) (1.470) (0.656) (0.900)
Teuro 0.035 * 0.045 ** 0.072 *** 0.036 * 0.042 ***
(0.015) (0.016) (0.027) (0.019) (0.015)
Euro 0.014 -0.008 0.052 -0.007 0.005
(0.023) (0.027) (0.041) (0.026) (0.025)
[D.sub. 2.082 1.669 0.663 2.031 1.821
>2002] (1.632) (1.687) (2.410) (1.693) (1.740)
[pi] 0.517 0.664 0.508 0.174 1.137
(0.677) (0.692) (1.280) (0.650) (0.734)
Rising 0.160 * 0.186 * 0.236 0.189 * 0.138
inflation (0.073) (0.089) (0.163) (0.083) (0.094)
Falling -0.013 -0.038 -0.175 0.007 -0.029
inflation (0.076) (0.094) (0.162) (0.106) (0.093)
[D.sub. 3.237 *** 3.136 * 2.121 3.209 * 3.169 *
>2002] (1.178) (1.245) (1.891) (1.346) (1.260)
<2002 4.899 *** 5.587 ** 7.036 * 4.922 *** 6.547 ***
[pi] (1.375) (1.747) (3.187) (1.604) (1.761)
Rising 0.052 0.052 0.020 0.012 0.042
inflation (0.099) (0.102) (0.206) (0.102) (0.118)
Falling -0.037 -0.102 -0.199 -0.077 -0.009
inflation (0.098) (0.107) (0.192) (0.124) (0.110)
[greater 0.328 1.295 0.118 0.227 1.642
than or (1.255) (1.290) (2.308) (0.924) (1.461)
equal to]
2002
[pi]
Rising 0.160 * 0.220 *** 0.365 ** 0.228 ** 0.119
inflation (0.088) (0.072) (0.124) (0.064) (0.106)
Falling 0.133 0.136 -0.137 0.251 0.1 19
inflation (0.130) (0.151) (0.284) (0.172) (0.149)
Variable (11) (12) (13) (14)
ag3 ag4 mal fem
[pi] 1.976 * 1.504 ** 1.847 ** 1.822 *
(0.779) (0.737) (0.693) (0.748)
[D.sub. 1.345 1.065 1.012 1.109
>2002] (1.188) (1.379) (1.166) (1.282)
[pi] 2.887 ** 2.224 ** 2.591 *** 2.494 **
(0.804) (0.804) (0.713) (0.802)
Teuro 0.054 ** 0.029 * 0.038 ** 0.040 **
(0.019) (0.014) (0.014) (0.015)
Euro 0.021 -0.010 -0.000 0.007
(0.028) (0.035) (0.033) (0.026)
[D.sub. 2.416 1.266 1.909 1.623
>2002] (1.562) (1.992) (1.642) (1.633)
[pi] 0.321 0.249 0.347 0.768
(0.742) (0.851) (0.647) (0.650)
Rising 0.219 ** 0.151 * 0.192 * 0.155 *
inflation (0.077) (0.089) (0.081) (0.089)
Falling -0.065 -0.107 -0.078 -0.015
inflation (0.088) (0.114) (0.081) (0.095)
[D.sub. 3.665 *** 2.648 * 3.077 *** 2.994 *
>2002] (1.205) (1.466) (1.156) (1.234)
<2002 5.060 ** 4.109 * 4.513 *** 5.868 ***
[pi] (1.758) (2.233) (1.520) (1.713)
Rising 0.115 -0.009 0.061 0.054
inflation (0.098) (0.103) (0.094) (0.120)
Falling -0.146 -0.195 -0.151 0.004
inflation (0.122) (0.139) (0.095) (0.122)
[greater 0.323 0.685 0.946 0.832
than or (1.298) (1.092) (1.064) (1.142)
equal to]
2002
[pi]
Rising 0.255 ** 0.222 ** 0.246 ** 0.144 *
inflation (0.075) (0.080) (0.071) (0.072)
Falling 0.032 0.009 0.029 0.150
inflation (0.131) (0.176) (0.116) (0.170)
Notes'. 'Tot' stands for total, 're1-re4' distinguishes
consumers by household income: ret = 1st quartile, re2 = 2nd
quartile, re3 = 3rd quartile, re4 = 4th quartile. 'ed1-ed3'
distinguishes consumers by educational level: ed1 = Primary
education, ed2 = Secondary education, ed3 = Further
education. ag1-ag4 distinguishes respondents by their age:
ag1 = 16-29, ag2 = 30-49, ag3 = 50-64, age4 = 65+. 'mal' and
'fern' distinguish respondents by gender. Note that the
autoregressive parameter, inflation expectations, and a
constant are always included in the regressions but not
reported in order to keep the table tractable.
TABLE 4
Linearity Tests
Transition F F4 F3 F2 Suggested
Variable Model
[[pi].sup.perc 0.458 0.517 0.761 0.144 Linear
.sub.t-1]
[[pi].sub.t] 0.383 0.130 0.536 0.543 Linear
Volume (t - 1) 0.373 0.133 0.632 0.431 Linear
TREND * 0.017 0.990 0.003 0.262 LSTR2
Note: This table presents linearity tests for the LSTR
models. They were obtained using the codes in JMulti from
Helmut Lutkepohl and Markus Kratzig.
TABLE 5
LSTR Results
Start Estimate SD
Linear part
Const. 3.90 10.50 1.36
[[pi].sub.perc.sup.t-1] 0.10 -2.15 0.39
[[pi].sub.t] 0.03 0.10 0.06
[Volume.sub.t-1] 2.53 10.17 2.63
Nonlinear part
Const -3.21 -10.32 1.37
[[pi].sub.perc.sup.t-1] 0.31 2.82 0.42
[Volume.sub.t-1] -2.38 -9.64 2.69
[gamma] 10 61.56 12.35
[c.sup.1] 52.17 53.81 5.47
[c.sup.2] 52.17 53.81 5.47
[R.sup.2] .89
Adj [R.sup.2] .89
t-Stat p Value
Linear part
Const. 7.74 0
[[pi].sub.perc.sup.t-1] -5.50 0
[[pi].sub.t] 1.82 .07
[Volume.sub.t-1] 3.87 0
Nonlinear part
Const -7.53 0
[[pi].sub.perc.sup.t-1] 6.72 0
[Volume.sub.t-1] -3.58 0
[gamma] - -
[c.sup.1] - -
[c.sup.2] - -
[R.sup.2]
Adj [R.sup.2]
Note: This table presents coefficients and parameter estimates
for the LSTR models. They were obtained using the
codes in JMulti from Helmut Lutkepohl and Markus Kratzig.
TABLE 6
Linearity Tests with Tone Variables
Transition Suggested
Variable F F4 F3 F2 Model
[[pi].sup.
perc.sub.t-1] 0.175 0.024 0.071 0.108 Linear
[[pi].sub.t-1] 0.852 0.420 0.823 0.802 Linear
Volume (t- 1) 0.052 0.021 0.222 0.524 Linear
VoIumeNeut (t- 1) 0.479 0.197 0.510 0.691 Linear
Falling Infl (t - 1) 0.940 0.573 0.749 0.970 Linear
RisingInfl (t - 1) 0.131 0.128 0.146 0.457 Linear
TREND 0.189 0.425 0.137 0.260 Linear
Note: This table presents linearity tests for the LSTR
models. They were obtained using the codes in JMulti from
Helmut Lutkepohl and Markus Kratzig.
