Public preferences about macroeconomic priorities are an essential
input for democracies into any explanation of the choice of monetary
institutions, the extent of international monetary cooperation, the
adoption of macroeconomic policies, and the robustness of economic
performance. Whether the objective is explaining institutional choice,
international coordination, policy choice, outcomes, or the associated
political conflict over these decisions, political economy models must
specify the preferences of all the relevant actors. Although the
preferences of politicians, central bankers, and even interest groups
are often discussed in these models, the investigation of these
research questions has largely taken place without reference to what
the public thinks about economic policy objectives. When the
preferences of citizens have been included in this research, those
preferences are usually assumed rather than grounded in empirical
evidence. What are the macroeconomic priorities of individual citizens?
Are views about macroeconomic priorities similar across different
individuals, or are there important divisions? Do preferences vary
across countries, and, if so, what accounts for this variation?
This article uses individual-level survey data from twenty advanced
economies to address these questions. There are three main empirical
results. First, the findings suggest that economic context, defined by
inflation and unemployment performance, has a substantial impact on the
public's economic objectives in a way broadly consistent with the
specification of utility/loss functions in the theoretical
political economy literature. Rising and more volatile inflation is
more costly, and the public places greater emphasis on low inflation as
prices increase more rapidly. Similarly, as unemployment rises,
reducing unemployment becomes a greater priority. These results are
generally consistent with a very large public opinion literature,
though this study extends those results by relying on comparable data
from twenty advanced economies.
Second, the findings in this article suggest that there is
significant cross-country variation in inflation aversion, controlling
for economic context. The mass public's assessment of the relative
costs of inflation and unemployment varies substantially across
countries. The third empirical result is that some of this
cross-national variation is accounted for by national-level factors
affecting the aggregate costs of inflation and unemployment. The
empirical estimates in this article suggest that the demand for
government revenue and the size and structure of the financial sector
partially explain cross-country variation in inflation aversion,
controlling for economic context.
Cross-country variation in inflation aversion while controlling for
economic context has significant implications for optimal monetary
policymaking, for explanations of variation in economic outcomes, and
for theoretical political economy models of alternative institutional
frameworks for policymaking. In standard macroeconomic models based on
rational expectations, optimal policy—whether chosen period by
period or by precommitment to an optimal rule—depends on levels
of inflation aversion in society. In particular, the optimal policy
will stabilize output in response to shocks more aggressively the lower
inflation aversion is in society. The cross-country variation in
inflation aversion documented in this article suggests that optimal
policy responses will differ across countries simply as a function of
preferences. This finding implies that both prescriptions for policy
and explanatory models of variation in macroeconomic outcomes should
take into account country differences in inflation aversion. The
implications for the adoption of alternative institutions for monetary
policymaking are also straightforward. For example, economic arguments
about joining monetary unions are often organized around determining if
the countries in question constitute an optimal currency area. This
evaluation has focused a great deal of attention on the degree to which
economic activity in the parts of the union is correlated, because one
of the major costs to currency unions is giving up the ability to use
the exchange rate and monetary policy to stabilize output and
employment. The empirical results in this article suggest that
assessments of the costs and benefits of currency unions must also take
into account variation in inflation aversion, as these differences
affect the costs and benefits to all members adopting a single policy.
Moreover, cross-country differences in inflation aversion will generate
conflict when the members of currency unions select the policymaker(s)
for the union. How conservative the optimal central banker is depends
on the levels of inflation aversion in society. Variation in those
preferences across the members of the currency union will generate
ongoing conflict about who best fills the role of the central
banker.
The remainder of the article is organized into six sections. The next
section defines inflation aversion and the following section provides a
theoretical framework for explaining its variation across individuals
and countries. The next two sections discuss the data and model
specifications and present the empirical results. The penultimate
section discusses some of the theoretical and methodological
implications of the findings, and the final section concludes.
Defining Inflation Aversion
The substantial literature on the political economy of macroeconomic
policymaking provides a theoretical structure for the analysis of
variation in macroeconomic priorities.1
In these models, voters and policymakers are
assumed to have utility or loss functions that depend on inflation and
output or unemployment. The exact functional form of the utility or
loss functions varies across different contributions to the literature,
but the main intuition is that utility decreases in the inflation rate
and the unemployment rate and increases in the level of output.
Typically, only a single indicator of real economic
performance—unemployment or output growth—is included in
any particular specification of the utility or loss functions.
2 Given the macroeconomic models employed in
the literature, the choice of indicators of real economic activity does
not matter for key theoretical results, although in richer models of
the economy it could, of course, make a difference. This article
follows much of the literature by focusing attention on unemployment as
the relevant indicator of real economic activity, in part because it is
the most salient measure in public discussions of the economy.
A common specification, the Barro-Gordon loss function, is the
following equation:
3 where Ut is the employment rate;
Utn is the natural
rate of unemployment; πt is the inflation rate;
a and b indicate the relative weight of the first and
second terms in the loss function and are assumed to be strictly
greater than 0; and k indicates the extent of distortions,
such as unemployment compensation and income taxation, that make the
natural rate exceed the efficient or socially optimal rate and is
generally assumed to be less than 1 but greater than or equal to 0.
This loss function values price stability and full employment, and the
ratio of the parameters a and b captures the benefit
of employment relative to the cost of higher inflation. These
parameters therefore indicate the voter's or policymaker's
inflation aversion—that is, how the individual assesses the
relative costs and benefits of inflation and unemployment. The subject
of this article is an empirical investigation into whether inflation
aversion varies across countries, and if so, what accounts for that
variation.4
The loss function in equation
(1) is actually symmetric for unemployment, with departures from the
natural rate in either direction generating losses. Differences in both
directions are weighed equally for the special case in which k
= 1. Consequently, to say voters like full employment means they like
unemployment equal to the natural rate. In most of the theoretical
political economy literature, k is assumed to be less than one
(for example, the existence of a time consistency problem in monetary
policymaking depends on k being less than one), and it is
accurate to say voters especially dislike outcomes in which
unemployment is high.
What Accounts for Cross-Country Variation
in Inflation Aversion?
This section discusses the theoretical determinants of inflation
aversion. If there is cross-country variation in the public's
macroeconomic priorities, what are its determinants? The explanatory
framework focuses on those factors that affect perceived costs of
inflation and unemployment at both the individual and national
level.
How individuals assess the importance of alternative macroeconomic
policy objectives should obviously depend on current economic
performance. Because the costs of inflation increase in the inflation
rate and the costs of unemployment increase in the unemployment rate,
public concern about these economic outcomes will vary accordingly.
Consistent with these observations, a large body of public opinion
research has shown that individuals dislike unemployment, low growth,
and inflation.5
Note, however, that the responsiveness of public
concern about inflation and unemployment to current performance does
not reflect variation in inflation aversion as defined in the previous
section. The Barro-Gordon loss function in equation (1) and similar
utility/loss functions in the literature include unemployment and
inflation as arguments in the function. Inflation aversion
itself—assessments of the relative costs and benefits of
inflation and unemployment—measures how a given level of current
unemployment and inflation affects individual utility. In discussing
the determinants of variation in inflation aversion across individuals
and countries, one needs to know why, for a given level of inflation
and unemployment, some individuals and countries seem to have different
levels of concern about these economic objectives.
6 The Barro-Gordon loss function in equation (1) suggests
that economic context may not be a simple function of current inflation
and unemployment. It is possible that it is only departures of
unemployment from the natural rate that are perceived to be losses
(this would be true if the parameter k = 1). This would
require individuals to be fairly sophisticated, in that their
preference for employment, and growth is only for those levels
consistent with stable prices. Moreover, Barro and
Gordon 1983b and others have typically assumed that because of
existing labor market distortions, the natural rate is above its
efficient or socially optimal level. Consequently, when the natural
rate is higher, the costs of unemployment are higher, and so this may
be a relevant consideration as individuals set their macroeconomic
objectives. These considerations suggest that the natural rate of
unemployment may be a determinant of the economic context that affects
individuals' concerns about inflation and unemployment. This
possibility is addressed in some of the empirical work described
below.
Individual-Level Determinants
One possible explanation for cross-national variation in inflation
aversion is differences in the composition of individuals within a
country. Previous research has found that the distributional
consequences of inflation and unemployment affect individual
macroeconomic priorities.7
If the proportion of
individuals relatively more exposed to the costs of inflation
(unemployment) is greater in one country than another, then average
inflation aversion in that country can be expected to be higher
(lower).
Previous research suggests that individuals differ in their exposure
to the distributional consequences of inflation and unemployment
through both their position in the labor market and their ownership of
nominal assets and liabilities. The effect of labor market position is
straightforward—individuals vary in their risk of unemployment.
Higher national unemployment typically is not distributed equally
across labor market participants. For example, all else being equal,
low-income individuals have a higher probability of unemployment.
Moreover, differences by income group in concern about unemployment and
inflation may also be because of wage effects. In some countries,
periods of high unemployment have been associated with increasing wage
and income inequality.8
Consequently, lower-income groups may be
relatively more concerned about unemployment, generating a positive
relationship between income and inflation aversion.
9 Alternative hypotheses about the relationship between
income and inflation aversion are prevalent. It may be the case that
inflation is correlated with rising/high income inequality as the
rich find it easier than the poor to protect themselves against real
income declines from inflation. This argument suggests a possible
negative relationship between income and the demand for low inflation.
Alternatively, the relationship may be nonlinear, with low- and
high-income individuals being most concerned about inflation and
middle-income citizens being most concerned about unemployment. See
Alt 1979.
The more general point is
this: how national unemployment levels are likely to affect
individuals' labor market outcomes will differ across individuals,
and these differences will affect individual macroeconomic
priorities.
10 Given that risks of
unemployment may vary across individuals according to other demographic
characteristics besides income, such as education and gender, the
empirical analysis will investigate individual variation in inflation
aversion with respect to these characteristics in addition to income
and, of course, current employment status.
Besides distributional consequences in the labor market, it has long
been recognized theoretically that the relative concern of individuals
about inflation and unemployment may depend on the ownership of nominal
assets and liabilities. Unexpected inflation redistributes wealth and
income through all nominal assets and liabilities when contracts are
less than fully indexed. Debtors and those who pay fixed nominal
incomes gain at the expense of creditors and those who receive fixed
incomes. These redistributional effects of inflation are likely to
affect the macroeconomic priorities of individuals. Specifically, those
individuals who own nominal assets and receive fixed incomes are likely
to be more inflation-averse, while those who are debtors and pay fixed
incomes are likely to be less so.11
Scheve 2003 shows strong evidence that nominal
asset owners are relatively more inflation-averse in the United
Kingdom, but finds no evidence that those with nominal liabilities are
less so.
In addition to, or perhaps as part of, the distributional
explanations for variation in individual inflation aversion, the
macroeconomic priorities of individuals are often thought to reflect
the political coalitions with which they identify. There is substantial
evidence that parties of the left and right have systematically
different preferences about macroeconomic objectives, with parties of
the right being more inflation-averse than parties of the left.12
Consistent with this literature,
those individuals who identify themselves as ideological conservatives
or as members of right parties may be relatively more inflation-averse.
The hypothesized relationship must, however, be interpreted with some
care. It is unclear to what extent political ideology explains
macroeconomic preferences. Ideology and partisanship are sometimes
thought to be the result of early socialization (particularly the
influence of parent ideology on children), and thus it would be
accurate to say ideology and partisanship explain preferences. Other
accounts see these variables as summary statistics for policy
preferences or running tallies of political preferences that are
constantly updated. From this perspective, it is not clear that any
correlation between political ideology and measures of inflation
aversion indicates evidence of an explanation of individual
macroeconomic priorities.
The key question asked in this article about individual differences
in inflation aversion is whether the characteristics that account for
these differences also explain variation across countries. This is, of
course, only possible if there is substantial variation across
countries in the composition of citizens with the relevant
characteristics.
National-Level Determinants
Another possible explanation of cross-national differences in
inflation aversion concerns characteristics of the national economic
and political environment that affect the actual costs of inflation and
unemployment. If inflation (unemployment) is simply more costly to all
individuals in a particular country, then average inflation aversion
can be expected to be higher (lower).
One such country-level characteristic is based on variation in the
demand for government revenue. Current research on the costs of
inflation focuses attention on the effects of the interaction between
inflation and the tax system on welfare. This literature shows that
this interaction can have substantial negative effects.13
These negative effects are balanced against the
recognition that inflationary finance of government expenditures can
have some benefits. If nondistortionary lump sum taxes are not
available, then raising revenue through the inflation tax may be better
than other forms of taxation that distort economic behavior. In this
scenario, inflation is still costly, and these costs will limit the
extent to which it should be used as a source of revenue. Nonetheless,
there are potential benefits to inflation given that existing tax
structures are distortionary. How significant this consideration is in
evaluating the relative costs of inflation and unemployment is likely
to depend on the demand for government revenue. Suppose this demand is
assumed to be exogenous, depending on tastes for public services and
current or past military needs. Two reasonable indicators of this
demand are total government expenditures and total government debt.
This suggests that individuals in countries with higher government
expenditures and debt are likely to be, all else being equal, less
inflation-averse.
Public assessments of the relative costs of inflation and
unemployment may also depend on how open the national economy is to
international trade. The general intuition in the literature is that
inflation is more costly in more open economies.14
The reasoning behind
this hypothesis differs across contributions to this literature.
However, the intuition that higher inflation is correlated with greater
uncertainty about future inflation and greater real exchange-rate
volatility, which is more costly in more open economies, is sufficient
for this analysis. If so, individuals in more open economies may be
more inflation-averse.
Another salient argument in the literature on the determinants of
inflation outcomes is that the size and structure of the financial
sector is an important factor affecting the choice of monetary
institutions, policy, and ultimately economic outcomes.15
The claim is that the financial sector,
particularly firms engaged in traditional commercial lending with
typically long-term assets and short-term liabilities, has a strong
preference for price stability.
16 Some
financial sector business activities, such as foreign exchange trading,
may benefit from higher and more volatile inflation, so the emphasis
here is on traditional commercial banking businesses.
Strictly
interpreted, this argument might apply only to the distribution of
preferences about macroeconomic priorities— that is, it implies
that the financial sector is more inflation-averse than the rest of
society. However, a large financial sector might affect average levels
of inflation aversion at the margin, both directly through individuals
employed in the sector and indirectly through the sector's
influence on the media.
Finally, cross-national variation in inflation aversion may be a
function of differences in historical economic performance. Two
contradictory arguments are made with respect to the impact of national
historical economic performance on mass attitudes about inflation. The
most common argument is that in those countries that have experienced
significant inflationary periods, individuals are more concerned about
the potential of rising prices to foster economic and political
instability. In such countries, there is expected to be generally
greater demand for low inflation. One reason for this effect may simply
be pure learning on the part of the public and policymakers.
Alternatively, this effect may be due at least partly to constructed
and politically contested interpretations in which influential ideas
help shape the public's view of historical economic performance
and thus how that experience informs assessments about the relative
costs and benefits of inflation and unemployment.17
See Hall 1993; Hayo 1998; and McNamara
1999. More generally, differences in public information sets in
particular countries may lead individuals to have different priorities.
Historical economic performance is just one possible source of
variation in public information. The media, professional economists,
and political elites may also have an effect. See Shiller 1997.
The alternative argument is
that in countries that have had substantial inflation in the past, ways
of reducing the costs associated with rising prices have been
developed, and thus individuals in those countries are generally less
concerned about it. Given these two sets of considerations, the net
effect of historical economic performance is likely to differ across
cases as the relative magnitude of the two contradictory effects of
experience with inflation is not clear and may vary across time and
countries.
18 The discussion in this
section has not exhausted the national-level factors that may affect
inflation aversion. One interesting extension that is left for future
research is how particular combinations of wage-bargaining institutions
and monetary policy institutions may condition opinion formation about
macroeconomic priorities. See, for example, Garrett
and Way 1999; Iversen 1999; and Franzese 2002. The results reported in this article
are robust to including these variables, but these unreported
regression specifications must be viewed with some skepticism. As
argued in the penultimate section of this article, monetary frameworks
should be viewed as consequences of, among other things, national
levels of inflation aversion. This does not mean that there cannot be
feedback effects from institutions to public levels of inflation
aversion, but it does mean that to estimate such an effect requires
valid instrumental variables and/or perhaps more data over time
than was available in this study.
In summary, the observation that citizens generally dislike inflation
and unemployment does not imply that their assessments of the relative
importance of these objectives do not vary in important ways. There are
substantial theoretical reasons to think that the costs of inflation
and unemployment differ across countries and individuals and thus
systematically affect the public's macroeconomic priorities.
Data and Econometric Model
To provide evidence of cross-country variation in public inflation
aversion and to evaluate individual and national-level determinants of
this variation, I examine survey-based measures of inflation aversion
in twenty advanced economies in selected years between 1976 and 1997.
The data for this analysis are from five cross-national surveys that
included respondents in Australia, Austria, Belgium, Canada, Denmark,
Finland, France, Germany, Greece, Ireland, Italy, Japan, the Netherlands,
New Zealand, Norway, Portugal, Spain, Sweden, the United Kingdom (UK),
and the United States. The surveys are Eurobarometers (EB) 5.0 (1976) and 48.0 (1997)
and the 1985, 1990, and 1996 waves of the International Social Survey Program
(ISSP).19
Each country is represented in at least one
of the five surveys, and one country, Germany, is included in all five. In
total, the data set comprises forty-four surveys in the twenty countries. The
EB and ISSP are cross-country surveys that collect information on the
current economic and political opinions of representative samples of
individuals in each country.
20 Note that
this article limits its analysis to large advanced industrial
democracies. Consequently, some available data from the EB and ISSP
surveys is excluded. Specifically, EB data from Luxemburg and ISSP data
from several transition and lesser-developed economies is not used in
the analysis. See the Appendix for further details about the
construction of the data set.
The theoretical dependent variable for the study is inflation
aversion. To measure this variable, I use responses to survey questions
about macroeconomic priorities that are generally in the form of the
following item:
What do you think the (NATIONAL) government should give greater
priority to, curbing inflation or reducing unemployment?21
This is the exact question wording from the
Eurobarometer 5.0. Each of the surveys asks respondents an equivalent
question. The exact question wording for each of the five surveys is
reported in the Appendix. The Appendix reports how the slight
variations in the wording of each of the questions across the five
surveys was dealt with, and how the statistical analyses take into
account additional uncertainty generated by these
differences.
I constructed the variable inflation priority equal to 1 if
the respondent gave the “inflation” response and 0 if he or
she gave the “unemployment” response. This question
requires respondents to reveal explicitly how important they think low
inflation is relative to the problem of unemployment. In this sense,
the measure is consistent with the inflation aversion parameters
a and b specified in equation (1) above. Inflation
aversion is increasing in the inflation priority variable.
The key criterion in assessing whether this is a good measure of
inflation aversion, as defined above, is if responses to the question
will be sensitive to individuals' assessments of the relative
costs and benefits of inflation and unemployment. This expectation
seems at least ex ante reasonable.22
Note that consistent with the literature, the
objective is to measure preferences about inflation and unemployment.
These preferences are of interest for, and in fact part of, the utility
functions in standard economic models for which there is no long-run
trade-off between inflation and unemployment.
It must be recognized, however, that individual responses to this
question depend on the economic context in which the question is asked.
Consequently, answers to this question, taken in isolation, can be
thought of as eliciting the individual's context-specific
inflation aversion. Just as utility in the theoretical literature
depends on current inflation and unemployment rates, answers to this
question will depend on the same factors. Inflation aversion
itself—assessments of the relative costs and benefits of
inflation and unemployment—is therefore measured by responses to
the question controlling for the current economic context, as will be
done in the analysis below.23
There are
at least three alternative strategies for measuring inflation aversion.
The first is to ask individuals survey questions specifically about
inflation without reference to other macroeconomic policy objectives.
The major disadvantage of this approach is that there is no budget
constraint or price explicit in the question. These sorts of questions
often fail to reveal much information, as variation in responses is
low. The survey evidence suggests that most people can be expected to
think prices should be kept under control, even if they disagree
strongly about the relative importance of various economic policy
objectives. A variant of this approach uses survey questions about the
importance of inflation relative to noneconomic values. See, for
example, Hayo 1998. This type of question has
a price, but not one that maps to the definition of inflation aversion
described above. The second approach is to measure the sensitivity of
government popularity to inflation performance. See, for example, Hibbs, Rivers, and Vasilatos 1982. While this
method avoids problems with the wording in survey questions, the
relationship between government popularity and inflation depends on
each country's political and economic institutions, and this
variation makes it extremely difficult to construct comparable measures
across countries. The third alternative, implemented by Di Tella, MacCulloch, and Oswald 2001, is to
estimate the sensitivity of individuals' reported
“happiness” or “life satisfaction” to inflation
and unemployment. This approach also avoids some problems with the
wording in survey questions (but see King, Murray,
Salomon, and Tandon forthcoming) and is well suited to producing a
single estimate of how inflation and unemployment affect welfare. However,
this approach does not allow analysis of variation in macroeconomic priorities
across individuals, which is central to this article. Nonetheless, as Di Tella, MacCulloch, and Oswald 2001 indicate, the
approaches are complementary.
The analysis investigates the individual- and national-level
determinants of individual macroeconomic priorities, as measured by the
inflation priority variable, across the twenty advanced
economies included in the data set. The theoretical discussion
emphasizes that how important individuals think low inflation is should
vary with current economic performance, defined by inflation and
unemployment. The inflation variable is the annual percentage
change in the consumer price index as published in the World
Bank's World Development Indicators.24
The
unemployment variable is equal to the current unemployment
rate as reported in the
OECD Statistical Compendium.
25 The
expectation is that the
inflation variable will, all else being
equal, be positively correlated with the dependent variable
inflation
priority, while
unemployment will be negatively correlated with
inflation priority.
The theoretical discussion also noted that individuals are expected
to vary in inflation aversion as a consequence of characteristics that
affect the perceived costs of inflation and unemployment at the
individual level. Most importantly, variation among individuals is
expected to be a function of the distributive effects of inflation and
unemployment. Individual variation may, if the distribution of the
relevant individual characteristics varies enough across countries,
account for cross-national differences in inflation aversion.
To test these hypotheses about the role of individual
characteristics, I constructed measures of income, labor force status,
age, sex, education, and political ideology. The variable income
quartile ranges from 1 to 4 and indicates whether the
respondent's income is in the first, second, third, or fourth
quartile of the income distribution for the respondent's country.
Because low-income individuals are generally thought to have a higher
probability of unemployment, and because high unemployment has been
associated with increasing wage inequality in some countries, I
hypothesize that income quartile will be positively correlated
with inflation aversion. unemployed is a dichotomous variable
equal to 1 if the respondent is currently unemployed and equal to 0 if
otherwise. This variable measures the individual's exposure to
unemployment based on his or her current labor force status.
Consequently, I expect unemployed to be, controlling for other
factors, negatively associated with the dependent variable
inflation priority.
The variable age equals the respondent's age in years,
and gender is a dichotomous variable equal to 1 for females
and 0 for males. education years is an ordered categorical
variable with nine categories corresponding to increasing years of
formal education. Differences in inflation aversion according to these
demographic characteristics are likely a function of distributive
effects not captured by the measures of income and labor market
position used in the analysis, and of systematic differences in how
various types of individuals perceive the relative costs of inflation
and unemployment. For example, older individuals are more likely to
have significant nominal assets and/or to rely on fixed incomes.
Moreover, individuals who have lived during periods of substantial,
disruptive episodes of inflation are more likely to believe that low
inflation is a critical policy objective than those who have not.
Similarly, women have more volatile employment experiences than men,
controlling for skill and experience, in many countries. These
differences may lead to variation in preferences about macroeconomic
priorities with women being less inflation-averse than men. Education
is a common measure of both labor market skills and cognitive
abilities. To the extent that it is a skill measure, education likely
has effects on inflation aversion similar to those of income.
Finally, two measures were constructed to test the hypothesis that
the political coalitions with which individuals identify may affect
their macroeconomic priorities. political ideology is a
10-point measure with 1 indicating that respondents placed themselves
at the far left of the left/right scale and 10 corresponding to the
far right. party is a 5-point scale constructed by first
determining what political party the respondent identified with and
then placing the party on a 5-point left/right scale. These
variables are hypothesized to be positively correlated with the
dependent variable inflation priority, consistent with the
literature on partisan patterns of economic policymaking.26
See Alesina 1987;
Alesina, Roubini, and Cohen 1997; and Hibbs 1987a and 1987b.
One set of hypotheses about individual variation in inflation aversion
that cannot be tested in these data is the impact of ownership of
nominal assets and liabilities. Scheve 2003
demonstrates a strong correlation between nominal asset ownership and
inflation aversion. None of the survey data in this article report
detailed information on respondents' nominal assets and
liabilities.
The theoretical discussion also suggests that the costs and benefits
of inflation may vary across countries according to national-level
political and economic characteristics. One such characteristic may be
the demand for government revenue. I constructed two indicators of this
demand. government spending is equal to total government
expenditures as a percentage of gross domestic product, and
debt is equal to total government debt, also as a percentage
of gross domestic product. Because one of the hypotheses of this
article is that increasing demand for government revenues increases the
incentives for inflationary finance as an alternative to distortionary
taxes, I expect that both these measures will be negatively associated
with the inflation priority variable.
To evaluate the hypothesis that inflation may be more costly in more
open economies and thus may lead the public to be more inflation-averse
in countries more exposed to trade, I constructed the variable
trade openness, equal to imports plus exports as a percentage
of gross domestic product. This variable is hypothesized to be
positively correlated with inflation priority. The theoretical
discussion also considered the possibility that the size and structure
of the financial sector may influence the public's perceptions of
the relative costs of inflation and unemployment. The variable
financial employment is equal to total employment in the
financial sector as a percentage of total civilian employment, and
broadly measures the size of this sector. I expect that this variable
will be positively associated with the measure of inflation aversion,
inflation priority.
Table 1 reports the summary statistics of
the inflation aversion measure and the explanatory variables. The
empirical work aims to see how different factors affect perceptions of
the costs and benefits of inflation and unemployment, and thus the
probability that an individual places priority on price stability. The
dependent variable inflation priority equals 1 when a
respondent gives the “inflation” response and 0 for the
“unemployment” response. Then E(inflation
priorityi) = Pr(inflation
priority = 1|ψi) =
ψi, where i indexes each individual
observation and ψi equals the probability that
an individual gives the “inflation” response. I model the
variation in ψi according to the logistic form
with ψi = 1/[1 +
exp(−xi β)]. In
this expression, xi is a vector of
explanatory variables hypothesized to effect the probability of placing
priority on inflation, and β is a vector of effect parameters. I
estimate these effect parameters using logit regressions. Further,
because some of the key independent variables are country-level
aggregate measures, it is important to consider the possibility that
disturbances will be correlated within countries. If the disturbances
are correlated within groupings that are used to merge aggregate data
with individual-level data, then standard errors from the usual maximum
likelihood estimation of the logit model can be seriously biased
downwards.27
See Moulton 1990 for a discussion of this effect for
ordinary least squares regression.
The standard errors
reported in this analysis adjust for clustering by country and require
the much weaker assumption that errors are independent across countries
but not necessarily across every survey respondent within a country.
The analysis is based on five models (Models 1 through 5) defined by
five alternative sets of explanatory variables. Model 1 includes the
measures of economic context, inflation, and
unemployment, as well as a dichotomous variable indicating
whether the respondent was queried in the 1980s, a similar indicator
for the 1990s, and a series of dichotomous variables for each country
in the analysis except the UK. Model 2 adds to this initial
specification individual variables measuring labor market exposure and
demographic characteristics. Models 3 and 4 add to this specification
the two different measures of the respondent's political
conservatism, political ideology and party,
respectively. Model 5 adds to Model 2 the variables debt,
government spending, trade openness, and financial
employment. Table 2 shows the
hypothesized sign for each of the regressors.
Hypothesized sign of independent variables
Empirical Results
The logit regression coefficient estimates for Models 1 through 5 are
reported in Tables 3, 4,
and 5. The results suggest that economic
context has a substantial impact on the public's
macroeconomic priorities in a way broadly consistent with the
specification of utility/loss functions in the theoretical
political economy literature. Rising and more volatile inflation is
more costly, and the public places greater emphasis on low inflation as
prices increase more rapidly. Similarly, as unemployment rises,
reducing unemployment becomes a greater priority. The analysis also
reveals that there is substantial cross-country variation in
macroeconomic priorities, controlling for economic context. The
findings indicate that national-level factors that likely affect the
relative costs of inflation and unemployment on national economic
welfare account for some of this variation. Further, variation in
individual characteristics that affect the distributive consequences of
inflation and unemployment and thus shape how individuals weigh
different economic objectives helps explain individual-level variation
in macroeconomic priorities but does not account for much of the
cross-country variation.
Determinants of inflation aversion: Model 1
Determinants of inflation aversion: Models 2–4
Determinants of inflation aversion: Model 5
For the Model 1 results reported in Table
3, the estimated coefficient for the variable inflation
is 0.064 with a standard error of 0.022. This indicates that increases
in inflation are positively correlated with the probability that a
respondent places priority on curbing inflation. Substantively, it
suggests that for the base case of a UK respondent, increasing the
inflation measure from one standard deviation below its mean to one
standard deviation above while holding the other variables at their
means, raises the probability of an “inflation” response by
12 percentage points. This estimate of a positive correlation between
inflation performance and the inflation priority variable is
replicated for Models 2 through 5.28
For
Model 5, where control variables are included that measure the costs of
inflation to national economic welfare, the estimate is, as expected,
slightly smaller in magnitude with a relatively larger standard error
(p-value of 0.14).
The finding is consistent with
both the inclusion of an inflation variable in the utility or loss
functions in theoretical political economy models and with the large
public opinion literature suggesting that the public dislikes
inflation.
The estimated coefficient for Model 1 for the variable
unemployment is −0.059 with a standard error of 0.026.
As expected, unemployment is negatively correlated with the probability
that a respondent places priority on inflation. The magnitude of this
effect is comparable to that of inflation. Again for the base case UK
respondent, increasing the unemployment variable from one standard
deviation below its mean to one standard deviation above while holding
the other variables constant at their means, lowers the probability of
an “inflation” response by 12 percentage points. This
estimate is also robust across Models 2 through 5 and consistent with
previous theoretical and empirical research that specifies the
macroeconomic preferences of the public.29
In alternative specifications, some combination of the
unemployment rate, the natural rate of unemployment, and/or the
difference between these was included in the model. The results from
these analyses broadly confirmed the correlation between real economic
performance and the macroeconomic priorities of citizens. These
analyses did not allow for reliable inferences about whether
citizens' macroeconomic priorities were consistent with a
sophisticated natural rate model. See Sekhon
1999 for such an analysis.
The more important results reported in Table
3 are the coefficients on the country indicator variables.
Recall that the UK is the omitted country, so the coefficient estimates
indicate whether, holding the other variables constant, inflation
aversion among citizens is on average higher or lower than in the UK.
The estimates range from −0.776 with a standard error of 0.170
for Sweden to 0.347 with a standard error 0.042 for Australia. This, of
course, suggests that controlling for economic performance, Swedish
citizens are significantly less inflation-averse than UK citizens, but
that Australians are significantly more so. Specifically, at the same
average level of economic performance, the average Swedish citizen is
estimated to have an expected probability of placing priority on
inflation of 0.23, compared to 0.40 for the UK and 0.48 for
Australia.
Note that only by controlling for economic context are comparisons of
national differences in the inflation priority measure
informative. The estimates allow a rough empirical test of whether it
is reasonable to assume—as most political economy models
do—that inflation aversion is constant across countries. Clearly,
it is not, and the pattern of the estimates of country dummy variables
is broadly consistent with many stylized descriptions of relative
inflation aversion in the mass publics of advanced economies. For
example, Italy and France seem, on average, to have lower inflation
aversion compared to Germany and the United States. There are, of
course, exceptions, and many more surveys over different points in the
business cycle would be necessary to arrive at a reliable ordering.
Nonetheless, the results in Table 3 suggest
that there is significant variation in national inflation aversion,
controlling for economic context. Inflation aversion, controlling for
economic context, is, in fact, what the parameters a and
b in the Barro-Gordon loss function (and similar variations)
refer to. Consequently, the cross-country variation documented in Table 3 suggests that exchange-rate and currency
union political economy models which include multiple country loss
functions should allow for variation in preferences (parameters
a and b should be indexed by country) as well as
variation in industrial structure and economic institutions. The
penultimate section of the article offers a preliminary discussion of
what some of the consequences of country differences in inflation
aversion may be for standard political economy models.
The theoretical discussion has suggested that one possible
explanation for cross-country variation in inflation aversion is
differences in the composition of individuals within a country. The
coefficient estimates for Models 2, 3, and 4 reported in Table 4 suggest why this is a reasonable
possibility. These specifications add individual-level regressors to
determine if certain characteristics make individuals more or less
inflation-averse. The results in Table 4
indicate that labor market status, income, gender, and political
ideology are all correlated with individuals' macroeconomic
priorities.
For Model 2, the coefficient estimate for the income
quartile variable is 0.079 with a standard error of 0.016,
suggesting that higher-income individuals with fewer risks of actually
experiencing unemployment are generally more inflation-averse.30
This result is, of course, also consistent
with the related ideas that higher-income individuals are better able
to insure themselves against the risks of unemployment or that high
national unemployment levels may be associated with increasing income
inequality.
This result is not sensitive to changes in
conditioning variables and holds across Models 2 through 5. The impact
of unemployment risk is also evident in the estimated effect of being
unemployed. For Model 2, the estimated coefficient for
unemployed is −0.251 with a standard error of 0.035.
This finding is also robust across different specifications. These
results are consistent with the hypothesis that the distributive
consequences of economic performance—specifically individual
differences in labor market risks—affect macroeconomic
priorities. The results for the demographic variables are also of
interest. For Model 2, the estimated coefficient for the
gender variable is −0.237 with a standard error of
0.030. This indicates that women are, all else being equal, less likely
to place priority on curbing inflation than men. While this result is
open to interpretation, it may be a function of differences in labor
market risks by sex not measured by the other variables in the model.
The estimates for Model 2 also indicate that older respondents may be
relatively more inflation-averse, but that there is no significant
correlation between education and macroeconomic priorities.
31 Note that the estimate for age is
attenuated in Models 3 and 4, which add measures of political ideology
and partisanship. Because these variables may be, in part, consequences
of the demographic variables, this attenuation does not support the
inference that the age variable has no effect.
The
positive effect for age is consistent with expectations if older
individuals are more likely to own nominal assets and receive fixed
nominal incomes.
32 Recall that the data
do not allow directly testing for the impact of ownership of nominal
assets and liabilities on macroeconomic priorities. See Scheve 2003.
The final results reported in Table 4 that
merit attention are the estimates for the impact of political identity
on macroeconomic priorities. The estimates for both the political
ideology and party measures of political identity suggest
that conservatism is positively correlated with greater inflation
aversion: individuals who identify with political conservatism place
greater emphasis on price stability. As pointed out in the theoretical
discussion, this result is consistent with a central assumption in the
theoretical and empirical literature on partisan and rational partisan
business cycles.
Overall, the estimates in Table 4 make it
clear that differences in individual characteristics, particularly
those that indicate variation in exposure to the distributional
consequences of economic performance, affect respondents'
macroeconomic priorities. Whether these characteristics may account for
some of the cross-country variation in inflation aversion documented in
Table 3 is ambiguous. For this to be the
case, there would have to be sufficient differences across countries in
the relevant characteristics.
One way to evaluate this possible explanation is to calculate, using
the estimates from Model 2, the difference between the expected
probability of observing an “inflation” response when a
hypothetical respondent has values of the individual-level variables
equal to the sample averages and the expected probability when this
respondent has values equal to a specific country's averages,
holding economic context constant. Again focusing attention on Sweden,
the UK, and Australia, this difference is calculated to be 0.000,
0.002, and −0.005 respectively. This pattern, which holds
generally, indicates that while the individual-level variables
certainly help to explain variation in macroeconomic priorities across
individuals, there is simply not enough variation in these measures
across countries to account for much of the cross-national variation in
inflation aversion.
An alternative set of explanations for cross-country variation in
macroeconomic priorities involve characteristics of the national
economic and political environment that affect the actual costs of
inflation and unemployment. One such factor reviewed in the theoretical
discussion is the demand for government revenue. There are potential
economic benefits to inflation given that existing tax structures are
distortionary. How important a consideration this is in evaluating the
relative costs of inflation and unemployment is likely to depend on the
demand for government revenue. Table 5
reports estimates of the impact of the demand for government
revenue—as measured by the variables government spending
and debt—on the public's macroeconomic priorities.
The coefficient estimate for the government spending variable
is −0.072 with a standard error of 0.024, indicating a
significant negative correlation with the inflation aversion measure.
This result is consistent with the argument that in those countries
with greater revenue requirements, inflation is less costly and
individuals are less likely to place emphasis on price stability. The
coefficient estimate for the debt variable that was also
expected to measure the demand for government revenue, however, is not
significantly different from 0.33
One
reason that the debt result may be relatively small in magnitude is
that the analysis cannot control for the nominal assets held by
individuals. If high levels of national debt mean that citizens hold
more nominal assets, then this would attenuate the expected negative
effect based on the demand for revenue. The inflationary finance
interpretation of the strong correlation between government spending
and macroeconomic priorities is made cautiously. It is, of course,
possible that individuals in some countries simply have a stronger
taste for employment protection and thus prefer higher levels of
government spending and are less inflation-averse.
The Model 5 specification also includes a measure of trade openness.
The expectation is that inflation is more costly in more open
economies, and so individuals will weigh inflation more heavily. The
results are not, however, consistent with this hypothesis. The
estimated effect of increasing trade openness is negative and
statistically significant. Further investigation of the correlation
between trade openness and the inflation priority dependent
variable suggests that this negative correlation may be sensitive to
the choice of conditioning variables. Consequently, I simply interpret
the estimate as disconfirming the hypothesis of a positive
relationship.
The estimates in Table 5 for the
financial employment measure suggest that a large national
financial sector is associated with greater inflation aversion. The
coefficient estimate is 0.078 with a standard error of 0.044
(p-value of 0.073). The result provides an individual-level
mechanism that reinforces Posen's argument that the financial
sector, particularly firms in traditional commercial lending, has a
strong preference for price stability. It appears that a large
financial sector may affect average levels of inflation aversion,
perhaps directly through individuals employed in the sector and/or
indirectly through the sector's influence on the media.
Overall, these national-level factors affecting the relative welfare
costs of inflation and unemployment do have a systematic impact on
individuals' macroeconomic priorities. To evaluate whether these
factors account for cross-country variation, it is again instructive to
compare the expected probability of observing an
“inflation” response when a hypothetical respondent has
values of these national-level variables equal to the sample averages
and the expected probability when this respondent has values equal to a
specific country's averages, holding economic context and the
individual-level measures constant. Again focusing attention on Sweden,
the UK, and Australia, this difference is calculated to be 0.101,
0.018, and −0.296 respectively. This means that the expected
probability of observing an inflation response is 10 percentage points
higher for Sweden if one assumes values for debt,
government spending, financial employment, and
trade openness that are equal to the sample averages for these
variables rather than the actual averages for Sweden. There is
substantial cross-national variation in these variables, and they
contribute significantly in helping to explain variation in
macroeconomic priorities across countries.
Finally, some of the cross-country variation in inflation aversion
may also be because of differences in historical macroeconomic
performance. Although the arguments that “historical experience
matters” are often contradictory, it does seem likely that
national differences in public information sets, determined in part by
historical economic performance, play a role in the formation of
macroeconomic priorities. It may be possible in future research with
more data to assess explicitly the extent to which differences in
average inflation aversion across countries can be tied to historical
experiences.
The general idea that public information sets account for some of the
differences across countries and time not explained by other factors
receives some modest support in the logit regression estimates. Each
specification includes a dichotomous variable indicating whether the
respondent was queried in the 1980s and a similar indicator for the
1990s. The results indicate that, controlling for other factors such as
current economic performance, placing priority on price stability is
more likely for a respondent queried in the 1980s and 1990s than the
1970s.34
For Models 1–4, the
estimates are statistically significant at the .05 level for the 1990s
but not the 1980s. For Model 5, the estimate is significant at the .10
level for the 1980s and is significant at the .15 level for the
1990s.
It is entirely possible that these differences are
because of macroeconomic experiences with high and volatile inflation
in many of the countries in the sample during the 1970s and early
1980s. These historical experiences may inform current assessments of
the costs of inflation and thus the public's macroeconomic
priorities.
Theoretical Implications
Policymaking and Economic Outcomes
The first and most obvious implication of cross-country variation in
inflation aversion is that optimal monetary policies and economic
outcomes will vary across countries simply as a function of
preferences. To discuss this implication as well as those for
institutional choice and international monetary cooperation, it is
instructive to specify a simple model of macroeconomic
policymaking.35
The model discussed below
is a standard framework for analyses of the political economy of
macroeconomic policymaking. See, for example, Drazen
2000.
Assume that the economy is described by the following supply function:
where Ut equals the unemployment rate
relative to the natural rate of unemployment, assumed for convenience
to be equal to 0. The value of Ut in a
given period t depends on the difference between the actual
rate of inflation, πt, and the expected rate,
πte, and on a stochastic
unemployment shock, εt.
εt is independently and identically
distributed with mean 0 and variance σ2.
The policymaker is assumed to have a loss function in every period of
the general form of equation (1) simplified slightly for discussion
purposes. Let
where θ > 0 is the inflation aversion parameter indicating
the policymaker's assessment of the relative costs of inflation
and unemployment [equivalent to the ratio of b over
a in equation (1)]; and K is the socially
optimal unemployment rate [equivalent to the product of k
and Utn in equation
(1)]. As do Barro and Gordon, I assume that the policymaker
targets a level of unemployment less than the natural rate, because
some distortion makes the natural rate higher than what the policymaker
believes is socially optimal and therefore that K is negative
(recall that in this section the natural rate is normalized to 0). This
wedge is, of course, what generates the time consistency problem in
models of monetary policymaking.
To start, however, consider the hypothetical case in which the
policymaker is able to commit ex ante to an optimal rule. The
timing of policymaking is then that first the policymaker announces a
rule. Then, expectations,
πte, are formed assuming
the rule will be followed, and private agents commit to nominal
contracts. Then an economic shock εt is
realized and observed by the policymaker, who then sets
πt according to the rule.
The optimal policy is calculated by substituting equation (2) into
equation (3) and minimizing the expected value at t −
1(Et−1
Lt) subject to rational expectations under
the assumption that the rule will be followed
(πte =
Et−1πt). The
optimal policy rule is
Under this rule, the average inflation rate will be 0 and the average
unemployment rate will be at the natural rate—normalized to 0 in
this model. Policy, however, will achieve unemployment stabilization in
that it will respond to shocks to unemployment brought on by
technology, financial crises, and so on. Suppose it was possible to
implement such a rule—then the empirical results of this article
have important implications for optimal policymaking.36
This is not just a theoretical possibility,
as Blinder 1998 argues that central banks
target the natural rate of unemployment (K = 0) and so the
time consistency problem does not arise in practice.
The
analysis in the previous section suggests that θ, the inflation
aversion parameter, should be indexed across countries, as some
countries—for example, Germany and Australia—apparently
have higher values for this parameter than others—for example,
Sweden and Portugal. Consequently, the optimal policy response to
economic shocks differs across countries. Simply that Swedish citizens
care more about unemployment stabilization than Australian citizens
implies that policymakers in Sweden will maximize welfare by responding
more aggressively to a given economic shock than policymakers in
Australia. This result may be of particular importance for developing
countries that are not included in the empirical analysis. Assuming
that the existence of country heterogeneity in inflation aversion
generalizes to these countries, it is obviously poor advice for
international organizations and policy experts to recommend uniform
policy responses to similar economic events.
The assumption made in the foregoing analysis is that policymakers
are able to commit to the optimal rule. It is well known that this rule
is not time-consistent in that, if expectations were really formed
based on the rule, policymakers would have an incentive to depart from
it when they actually go to set policy (assuming K < 0 and
there actually is a wedge).
According to this view, the more reasonable set of assumptions for
the timing of policymaking are that first, inflation expectations,
πte, are formed. Next, an
economic shock εt is realized and observed by
the policymaker, and finally, the policymaker sets
πt.
To calculate the resulting policy under these assumptions, again
substitute equation (2) into equation (3). Then minimize with respect
to πt treating
πte as given. Then assume
rational expectations so that
πte =
Et−1πt. This
yields the equality πte =
(−1/θ)K. Substituting, the equilibrium
inflation rate is equal to
Under period-by-period policymaking, the average inflation rate is
(−1/θ)K and the average unemployment rate will
be at the natural rate. Again, policy will achieve some degree of
unemployment stabilization but no more so than under the optimal rule
discussed above. Thus the inability to commit to the optimal rule
decreases welfare, as inflation is higher with no compensating gains in
unemployment performance. For this model, the most important
implication of the article's empirical results is that variation
in inflation aversion will have a systematic effect on observed
inflation, because inflation aversion determines the degree of
inflationary bias. Thus empirical studies attempting to explain
variation in inflation performance need to somehow account for
cross-national variation in inflation aversion to avoid bias in their
estimations.37
This point assumes,
consistent with the model, that the public's welfare informs the
policymaker's welfare. In frameworks for which this is not the
case, as in a strong partisan account, the insight is not as telling
(assuming a valid measure of the inflation aversion of the partisan
group in power is employed).
It is also worth pointing out
that expressions for both inflation and output volatility in this model
are a function of inflation aversion {
Var(π) =
(1/[1 + θ])
2σ
2,
Var(
U) = (θ/[1 +
θ])
2σ
2}. Consequently, empirical
analyses that attempt to explain the volatility of inflation and
unemployment also need to account for cross-national variance in
inflation aversion (note that even for the optimal rule case, the
inflation aversion parameter matters for expectations about the
volatility of inflation and unemployment).
To illustrate the plausibility of the claim that inflation aversion
among mass publics may be a significant factor in explaining variation
in economic outcomes, I conducted a simple analysis evaluating the
relationship between country estimates of inflation aversion based on
the survey analysis in the previous section and average inflation
outcomes. Because all twenty countries included in the study have
surveys in the 1990s and because this decade makes up about 75 percent
of the total individual observations in the data set, I focus the
analysis of average inflation outcomes on the period 1990 to 1997.38
If instead average inflation outcomes for
the entire period of the survey data are analyzed (1976–97), the
results are qualitatively the same.
The variable
estimated
inflation aversion employs the parameter estimates from Model 1
reported in
Table 3, and is equal to the
predicted probability that a respondent from a particular country in
the 1990 decade gives the “inflation” response assuming
that the economic context (
inflation and
unemployment) is equal to the average for the entire sample.
In the analysis above, I have claimed that responses to the survey
question, controlling for economic context, measure inflation aversion.
Thus the predicted probability described above is a country-level
measure of inflation aversion (alternatively the estimated coefficients
on the country dummy variables for Model 1 could be used and all
results reported here are the same employing this measure). The
variable
average logged inflation is equal to the average of
the natural log of the change in the consumer price index for the years
1990 to 1997.
Figure 1 plots the variables average
logged inflation and estimated inflation aversion and
draws the relevant regression line. The graph reveals a negative
correlation indicating that those countries with greater inflation
aversion experienced lower average inflation outcomes. Table 6 reports the estimates for two
ordinary least squares regressions of average logged
inflation on estimated inflation aversion. The
results for Model 6 indicate that the observed negative correlation in
Figure 1 is statistically significant, while
the estimates for Model 7 suggest that the partial correlation is
robust to including a measure of central bank independence.39
The measure of central bank independence is
the index developed by Cukierman, Webb, and Neyapti
1992. I use Keefer and Stasavage's 2002
update of this index for the 1990s—specifically I use the measure for
1994 that is approximately the midpoint of the years in my analysis (their
data does not allow for a calculation of the full 1990–97 average). I
employ mean imputation for Portugal for which the central bank
independence measure is missing.
There are a number of obvious
limitations to this analysis. First, because the survey data are
limited to twenty countries and there is insufficient data to develop
time varying measures for each country, the analysis is based on a very
small number of cross-sectional observations. Second, and related, the
literature—employing larger data sets—has suggested a
number of other regressors besides central bank independence that
should be included in a more complete analysis. This point is perhaps
not as telling as it may seem, as many institutional variables may be a
consequence of inflation aversion and thus inappropriate control
variables for an analysis seeking to estimate the effect of inflation
aversion in the mass public on economic outcomes. Third, inflation
aversion may be in part a consequence of economic outcomes. This is
certainly possible, but if the idea is that experience with inflation
makes citizens more concerned about its costs, then the bias from
endogeneity results in a conservative estimate of the effect of
inflation aversion on average inflation in this analysis.
Determinants of average logged inflation 1990-1997: Models
6–7.
Average logged inflation by estimated inflation aversion
Of course, many other objections common to small N
cross-sectional analyses could be raised. The point, however, is that
common models of macroeconomic policymaking suggest that variation in
inflation aversion may result in variation in economic outcomes. The
survey analysis in this study demonstrates strong evidence of
cross-national variation in inflation aversion. Thus inflation aversion
in mass publics may be an important factor in explaining economic
outcomes, and there is some evidence in Figure
1 and Table 6 that this is the
case.
Institutional Choice
The second straightforward implication of cross-country variation in
inflation aversion is that even countries that face similar objective
economic environments may choose different macroeconomic policymaking
institutions. The classic comparison to illustrate this point is the
choice between policymaking under the discretionary, period-by-period
regime described above and policymaking following a simple, credible,
and easily monitored zero-inflation rule. The trade-off between these
policymaking institutions is, of course, that the former generates
inflation bias but some degree of stabilization, while the latter
avoids bias but fails to respond to economic shocks. By comparing
expected utility as defined in equation (3) under each institution
given its equilibrium inflation and unemployment rates, it can be shown
that the period-by-period regime is preferred if
Because σ2 is the variance of
εt, this result simply means that if
unemployment shocks are large enough relative to the parameters θ
and K that determine the extent of inflation bias, the
benefits of stabilization under the period-by-period regime outweigh
the costs of greater inflation. Previous discussions of discretion
versus simple rules have focused on how cross-country variations in
economic structure that affect exposure to economic shocks are likely
to influence this institutional choice. The empirical results
establishing cross-country variation in inflation aversion suggest an
additional consideration. The condition in equation (6) is easier to
satisfy as inflation aversion (θ) increases, because this increase
decreases the extent of inflation bias in the period-by-period
policymaking regime. So somewhat counterintuitively, it is, all else
being equal, less likely to be optimal for countries with higher
inflation aversion to adopt the simple zero-inflation rule than for
countries with lower levels of inflation aversion. It is beyond the
scope of this discussion to evaluate this prediction systematically,
and it is important to recognize that there are other institutional
alternatives to discretion and simple rules. Nonetheless, it is
interesting to note that the European Central Bank (ECB) has an
explicit mandate for price stability (essentially a zero-inflation
rule), while the U.S. Federal Reserve does not. The data presented in
Figure 1 indicate that ten of the eleven
countries for which the ECB controls policy have substantially lower
levels of inflation aversion than the United States.
Although a wide variety of analyses of the choice of monetary
institutions can similarly be informed by the consideration of
cross-national variation in inflation aversion,40
Candidates include the family of
“second-generation” models of monetary institutions that
consider decisions about CBI and exchange-rate pegs simultaneously. See
Bernhard, Broz, and Clark 2002.
the
final example sketched here is the case of a country joining a monetary
union. The dominant framework for examining this problem is the theory
of optimal currency areas (OCA). This literature focuses primarily on
identifying benefits to a country from joining a union such as reduced
exchange-rate risks and transaction costs for international businesses,
and on identifying costs such as foregoing the ability to use monetary
policy to respond to country-specific macroeconomic shocks. The
empirical results in this article demonstrating evidence of
cross-country variation in inflation aversion suggest that the
implications of country differences in preferences/objective
functions should also be considered central to analyses of a
country's decision to join a monetary union and to evaluations of
alternative policymaking frameworks for such unions.
Analogous to the evaluation above of expected utility under the
period-by-period regime and the simple zero-inflation rule, one can
compare a country's expected utility in and out of a monetary
union assuming that the union and country face the same policy problem
but have different preferences/objective functions [values of
the parameters K and θ in equation (3)] and respond
to different economic outcomes (inflation, unemployment, and
unemployment shocks for the entire currency area versus values for a
single country). Although a full analysis of the consequences of
variation in objective functions for a country's decision to join
a monetary union is beyond the scope of this article,41
a few key observations are
instructive. First, it can be shown that joining a monetary union can
be equivalent to the appointment of an independent conservative central
banker for those countries with relatively low levels of inflation
aversion. This effect, of course, has the same benefits (lower
inflation and inflation volatility) and costs (higher output
volatility) as the appointment of a conservative central banker.
Second, if countries are assumed to choose the inflation aversion of
the currency union's policymaking body, their preferences
regarding this characteristic of the policymaker will depend on
interactions between their levels of inflation aversion, the relative
variance of country-specific and currency area-wide unemployment
shocks, and the covariance between these shocks. For example, all else
being equal, the less positively correlated shocks are in the union,
the more conservative is the preferred currency union policymaker.
Third, if the choice of the inflation aversion of the currency union
policymaker is made at the same time as the decision to join the union,
then even under the most optimistic assumptions about the relative
variance and correlation of economic shocks, cross-national variation
in inflation aversion can make it such that there may be no level of
inflation aversion for the currency union policymaker that leaves all
potential entrants better off for having joined the union.
Conclusion
This article provides new evidence on the determinants of individual
macroeconomic priorities based on survey data in twenty advanced
economies. There are three main empirical results. First, the results
suggest that economic context has a substantial impact on the
public's economic objectives in a way broadly consistent with the
specification of utility/loss functions in the theoretical
political economy literature. Second, the results suggest that there is
significant cross-country variation in inflation aversion, controlling
for economic context. Third, some of this variation is accounted for by
national-level factors affecting the aggregate costs of inflation and
unemployment. The empirical estimates in this article suggest that the
demand for government revenue and the size and structure of the
financial sector partially explain cross-country variation in inflation
aversion, controlling for economic context and the individual
characteristics of survey respondents.
Overall, the findings in this article suggest a number of questions
for future research. As discussed in the previous section, the
existence of cross-country variation in inflation aversion, controlling
for economic context, has significant implications for optimal monetary
policymaking, for explanations of variation in economic outcomes, and
for models of alternative institutional frameworks for policymaking.
The foregoing discussion only sketched some of the most obvious
consequences, and further theoretical work could productively
incorporate the existence of country differences in macroeconomic
priorities into explanations of institutional choice, international
policy coordination, policy choice, outcomes, or the associated
political conflict over these decisions. The results in this article on
the role of individual characteristics in determining individual
macroeconomic priorities may also be useful for this literature,
particularly studies of why countries adopt the monetary institutions
that they do. The role of distributive conflict among groups in society
is central to debate about this question.42
The findings of this article
suggest that those distributive conflicts are evident in the electorate
as well as among firms in various sectors of the economy. The results
in the article may also be useful in future empirical studies
explaining variation in economic outcomes, particularly investigations
of the effect of monetary institutions. As various scholars have made
clear, evaluating the effect of these institutions depends on first
specifying preferences.
43 This article provides substantial
evidence that there is sufficient variation in public macroeconomic
priorities across countries for the specification of preferences to be
significantly improved by accounting for the relative inflation
aversion of citizens.
Appendix
Data Description
The surveys used in the analysis are Eurobarometers (EB) 5.0 (1976) and 48.0 (1997) and the
1985, 1990, and 1996 waves of the International Social Survey Program
(ISSP). The EB and ISSP are cross-country surveys that collect information
on the current economic and political opinions of representative samples of
individuals in each country. The sources for each country year used to
construct the dependent variable, inflation priority, are reported in
Table 7. For the EB 5.0 data, the English version of
the question is: “What do you think the national government should give
greater priority to, curbing inflation or reducing unemployment?”
Individuals were coded 1 if they gave the “inflation”
response and 0 if they gave the “unemployment” response.
Missing data was imputed using the procedures described in the next
section of the Appendix. The question for the EB 48.0 data differed in
the responses coded. The English version of the question is: “Do
you think the national government should give higher priority to
reducing inflation or higher priority to reducing unemployment?”.
Answers were coded on a 5-point scale ranging from “a lot higher
priority on reducing unemployment” to “a lot higher
priority on reducing inflation.” This scale was collapsed to a
dichotomy with “inflation” answers coded a 1 and
“unemployment” answers coded a 0. The dependent variable,
inflation priority, was coded using the same imputation and
coding rules described above. Finally, data from all three
International Social Survey Program (ISSP) studies are based on
responses to the question: “If the government had to choose
between keeping down inflation or keeping down unemployment, to which
do you think it should give highest priority?” Again, responses
indicating “inflation” were coded 1, and those indicating
“unemployment” were coded 0. The same procedures described
above were used to construct the dependent variable inflation
priority.
Methodology for Missing Data
The data constructed for this article are not fully observed. The
sources of the missing data range from the refusal of survey
respondents to answer particular items in some surveys to some
questions not being asked at all in certain surveys. Incomplete data,
whatever the source, can create a number of serious problems for making
valid statistical inferences. The most general and extensively
researched approach for dealing with missing data problems is
“multiple imputation.” King, Honaker, Joseph, and Scheve;
Schafer; and Rubin describe the advantages of multiple imputation over
alternative strategies for addressing missing data problems for survey
analyses like those presented in this article.44
The approach has several variations but always involves three main
steps. First, some algorithm is used to impute values for the missing
data. In this step, m (m > 1)
“complete” data sets are created consisting of all the
observed data and imputations for the missing values. The second step
simply involves analyzing each of the m data sets using
standard complete-data statistical methods. The final step combines the
parameter estimates and variances from the m complete-data
analyses to form a single set of parameter estimates and variances.
Importantly, this step systematically accounts for variation across the
m analyses due to missing data in addition to ordinary sample
variation.
The first step in the multiple-imputation procedures was to create
imputations in the missing data cells for all the variables used in the
analysis. The variables in the imputation model included all those used
in the analyses reported above as well as additional information from
the surveys determined to be helpful in predicting the missing data.
The imputation model included two similar measures of inflation
aversion. In addition to inflation priority, a three-category
ordinal variable was included that accounted for the fact that the
question wording in one of the five surveys discussed in the previous
section allowed for a neutral response. In constructing inflation
priority, these neutral responses were coded missing but then
imputed in the imputation stage of the analysis.45
See King, Honaker, Joseph, and Scheve
2001; and Schafer 1997 for discussion of why
it is optimal to impute the dependent variable as well as the independent
variables. The key intuition is that the goal of the imputation is to preserve
the observed distribution of the data. The imputations do this in a way
that allows all of the existing information in the observed data set to
be included in standard statistical analyses. The imputations
themselves add no new information.
All the results reported in
the text are robust to using either the
inflation priority
variable or the three-category ordinal variable to measure inflation
aversion. Altogether I imputed five complete individual-level data sets
for the pooled survey data. The exact imputation algorithm used is
known by the acronym “EMis” because to generate imputations
it combines a well-known Expectation Maximization missing data
algorithm with a round of importance sampling. King et al. provide a
complete explanation of the use of this algorithm for missing data
problems.
46 The imputation model was multivariate
normal with a slight ridge prior.
47 The
imputation procedures were implemented using Amelia: A Program for Missing
Data, Honaker et al., 2000.
The
final data sets contain completed observations equal to the actual
number of individuals in the survey. Also, all data sets contain the
same nonimputed information; they differ only in the imputations for
missing data.
The second step in the multiple-imputation analysis was to run the
various logit models separately on each of the five final data sets.
The last multiple-imputation step was to combine the five sets of
estimation results for each specification to obtain a single set of
estimated parameter means and variances. The single set of estimated
means is simply the arithmetic average of the five different estimation
results. The single set of estimated variances is more complicated than
a simple average because, as mentioned above, these variances account
for both the ordinary within-sample variation and the between-sample
variation due to missing data. See King et al. and Schafer for a
complete description of these variances.48
