Chapter 2

The Rational Expectations Hypothesis

as a Key Element of New Classical
Macroeconomics

Now, what I want is, Facts. Teach these boys and girls
nothing but Facts. Facts alone are wanted in life.
(Charles Dickens: Hard times)

2.1 Introduction

The contradictions and the reality of the underlying assumptions and conclusions of
the rational expectations hypothesis (REH) will be reviewed in this chapter. It will
be argued that crucial elements in new classical macroeconomics lead to confusion
in the argument, therefore unbiasedness needs to be reconsidered. Our critique is
based primarily on the fact that new classical macroeconomics is set to deliver a
realistic description of our life-world and, therefore, their models can be compared
directly to life-world conditions. However, it will be seen that the portrayal of a
typical agent as the REH postulated it is valid only as an ideal-type, that is, we will
have a problem unless this model is regarded as a pure theory. Although the degree
of rationality postulated by the REH cannot be a requirement under life-world
conditions, the rational expectations hypothesis is still not a futile theoretical
construct, since by finding a proper scope for this theory we can offer marvellous
theoretical support for the mechanisms of inflation targeting. If exaggerated
assumptions are set aside (that is, we ease the presuppositions) by which a monetary
authority capable of and destined for leading market expectations can be included,
we end up having a moderate version of the hard definition of the REH given by
Muth, while certain consequences of this concept will be consistent with the
assumptions of the original (hard) definition—assumptions that otherwise seem
indefensibly and groundlessly hard. [! experimental economics]
The outcome of talking about the easing of assumptions of macro-models (and
of all kinds of logical models actually) is dubious. Instead, one should, perhaps,
stress the incorrectness of these presumptions by underlining that an assumption
can be right or wrong (and nothing else) and that we can either maintain it or not.
Emphasizing the possibility of eased assumptions implies that assumptions that are
applied in various theories can be more or less correct. Is this not an absurd idea? It
might be useful to build our arguments upon an analogy. The central problem of
empirical studies investigating the validity of Barro-Ricardo equivalence was

© Springer International Publishing Switzerland 2015 53

P. Galbács, The Theory of New Classical Macroeconomics, Contributions to
Economics, DOI 10.1007/978-3-319-17578-2_2
54 2 The Rational Expectations Hypothesis as a Key Element of New Classical. . .

whether the fluctuations of consumption-related expenditures in the theory of

Keynes are, and to what extent, in accordance with actual processes (see
Chap. 5). A fairly flat pattern of consumption expenditures follows from the per-
manent income hypothesis; by contrast, a strong volatility follows from the Keynes-
ian theory. Every change in disposable income leads to proportional fluctuations of
consumption affected by marginal propensity to consume; while the permanent
income hypothesis predicts a much smoother time pattern. Whether consumer
expenditures fluctuate and to what extent is an empirical problem and, on the
very basis of empirical tests, they can be sufficiently described in quantitative
terms. Consumption expenditures may not fluctuate at all, they can be excessively
volatile or may show a considerable steadiness. If such qualifiers are not satisfac-
tory for a precise analysis (and they can hardly be so), we can say that the time
series data of consumption expenditures are more or less scattered around their
trend. Therefore, it seems that the predominance of a certain theory can be
measured in quantitative terms. Observed facts may be closer to the predictions
of the Keynesian or the monetarist theory; a clear case is expected to be only a
theoretical possibility. Anyway, what would a clear and striking manifestation of
the Keynesian consumption pattern stand for? Is there an economist who can
substantiate that an empirically estimated numerical value of the marginal propen-
sity to consume indicates the clear manifestation of the Keynesian theory? The
Keynesian consumption model and the permanent income hypothesis differ in the
implied numerical values of autonomous consumption and marginal propensity to
consume, since a higher exogenous component and a lower marginal term charac-
terize the latter. These two theories can be regarded as two extremities of possible
consumption patterns. But which values of the estimated parameters should be
labelled as clear manifestations of one theory rather than the other? It could be hard
to find a reasoned answer.
Therefore, in this sense, we can talk about the extent to which a theory pre-
dominates. In this way, easing and modifying certain assumptions of a model on a
continuum is not, therefore, possible? It is true, of course, that such easing leads to
another model, although it would be a transformed (constrained? modified?) ver-
sion of the original construct. In our view, assumptions can be eased and although
their literal predominance may be implausible; it is still possible, however, to find
an eased set that is (more) consistent with life-world conditions.
Assumptions of a model are not (or not necessarily) binomial variables. That is,
setting aside an assumption does not necessarily imply the contrary of the original
setting. Suppose that an assumption on the political environment is present in an
economic model. In this case, setting aside the assumption of a democratic scenery
does not imply a totalitarian dictatorship. Similarly, in the case of the market
system, if one does not use a monopolistic system as a starting point, this does
not result in the presumption of an absolutely free competition, since there are
plenty (in fact, innumerable) of intermediate structures between the two extremities
of free competition and monopoly. Therefore, it is not absolutely free competition
that comes after easing the rigour of a monopolistic environment but duopoly,
followed by multiple agent models of oligopolistic competition, all of which should
be regarded as a shift toward free competition.
2.1 Introduction 55

In his well-known methodological paper, Milton Friedman (1953) argues that if

one ends up having a theory consistent with reality in its predictions through
applying unrealistic assumptions, one should depart neither from our model nor
those assumptions.1 As it has already been mentioned above, Friedman’s example
is identifiable with the behaviour of leaves around a tree: by establishing a model in
demonstrating their density and thus finding a successful construct through com-
paring the predictions of some alternative models in the assumption that leaves are
rational agents capable of applying mathematic optimization methods—then this
model cannot (and should not) be contested. As an analogy, the geocentric and
heliocentric models of the Solar system come to mind. Both of these theoretical
constructs were successful in predicting the observed trajectory of the Sun in the
sky (cf. Weeks 1989). Therefore, in terms of the correctness of a model, it does not
matter whether it is capable of delivering consistent estimations and nothing else.
We do not (and cannot) discover the actual mechanism of reality, hence we cannot
explain a change in the consistency of the predictions if it does occur. As long as the
model works properly, its outputs are consistent with reality but not for the reason
the model itself suggests. We, critics, should not be satisfied with consistent pre-
dictions, since doing this would be tantamount to putting a stop to the development
of science, since if the only ground of our judgement is the consistency of pre-
dictions (and not the correctness of the model), to exceed the geocentric model
would not have been necessary at all.
Considering the lessons of Chap. 1, the assumption included by Friedman’s leaf
model is incorrect, since it stems from an arbitrary alchemy, not from abstraction.
However, it is excessive rather than incorrect to assume the existence of agents in
the rational expectations hypothesis, capable of establishing expectations consistent
with the objective probability distribution. The axiom is right: if, in actual fact,
economic agents had the methodological experience and the information-based
background (that is, if they were homo oeconomici) necessary for the establishment
of such expectations, nothing would prevent a particular theory from presenting a
realistic description of reality in synch with its underlying intentions.2 However,
human beings are not what the REH assumes them to be. By the way, this concept
may be accepted in terms of a pure theory and it should not even be criticized for
this: if Man was an agent created to be an analogy of the homo oeconomicus and an
actor in applications running on computer screens, this portrayal would be accept-
able. However, if the theory aims to describe reality realistically, this excessive
assumption cannot be maintained since it implies the immanent failure of the theory
itself (i.e. in its endeavour to offer a realistic description of reality). When referring
to the easing of the REH assumptions, a concept of agents might be used as a point
of departure (this, in fact, is the direction we have taken) who, rather than

1
At some point Friedman said that empirical tendencies alone prove unsatisfactory to suggest a
certain relationship between variables; clarification of the theoretical background is of paramount
importance (Friedman 1970).
2
However, it is always dubious whether theories should, in fact, paint a realistic and comprehen-
sive picture of reality. Is drawing a map on the scale of 1:1 sensible? (cf. Varian 1999)
56 2 The Rational Expectations Hypothesis as a Key Element of New Classical. . .

establishing their expectations independently, borrow them from a specialized

institution, where these expectations, which are not unbiased, are set to approximate
the objective probability distribution as much as possible. The distinction between
incorrect and excessive assumptions applied throughout this publication is based on
this very principle. The analysis is about a twofold exploration of assumptions: the
first question is whether an assumption is based on abstraction; the second is
whether the assumption itself is in accordance with the goals and ambitions of
the theory which the assumption is applied by. It should be emphasized that
assumptions are eased not for the purpose of turning erroneous presuppositions
into right ones—a wrong assumption cannot be fixed, only replaced (and replaced it
should be: the F-twist cannot be regarded as a legitimate scientific approach in
economics). However, intensifying and easing correct abstractions lead to model
families the analysis of which may enhance our knowledge.
The problem of whether the denial of new classical doctrines is based on the
irrelevancy or on the excessiveness of their assumptions, is difficult to address. We
will see in Chap. 3 that the general equilibrium analysis (general equilibrium
considered as a reference case) relies on the presumption of equilibrium. Simply
put, it is the axiomatic postulation of equilibrium. To claim that it is the most crucial
element of neoclassical theory is not an exaggeration. It is argued here that as an
abstraction, the assumption of tendencies pointing towards a state of equilibrium
should not be labelled incorrect: if changes in prices are not hindered, the state of
equilibrium will be established. This is the fundamental mechanism of market as
such.3 Nineteenth century neoclassical models—by distinguishing real-life circum-
stances from the model environment—applied the pure theory in a proper way (the
only danger was the notion of imperative referred to above). However, the general
equilibrium in the new classical theory as a presupposition is considered excessive
since the promise here was to provide a realistic description of reality. In other
words, there is a discrepancy between the purpose of the theory and the assumption
applied. However, to presuppose an equilibrium and equilibrium mechanisms is
still not unreal, since, after all, the abstraction that resulted in the axiomatic system
is correct. Were the opportunity to present itself, a state of equilibrium would in fact
emerge—the efforts to prove this are the greatest merits of mainstream economics.

2.2 Preliminary Remarks

It goes without saying that economists have been puzzled by the rational expecta-
tions hypothesis and the new classical macroeconomics by which it was created,
and that the scope of the theory itself still lacks clarification. Despite the fact that

3
Júlia Király (2000) claims that highlighting the equilibratory or non-equilibratory nature of
macroeconomic systems is a matter of faith and attitude. We are of the view that there is more
to it than that. General equilibrium models are abstraction-based constructions that grasp the
fundamental mechanisms of a purely economic setting through their ideal-types postulated as
axioms.
2.2 Preliminary Remarks 57

the REH, complete with the paradigm of continuous market clearing, underscores
the predestined failure of the activist fiscal and monetary policy,4 the theory has
become an integral part of modern economic policy recommendations (the theses
by Nobel awardees Finn E. Kydland and Edward C. Prescott should be mentioned
here). In pure logical terms, it can be clearly proven that rational expectations are of
crucial importance when explaining the mechanism of inflation targeting
(cf. Harasztosi 2004); or at least, it can be realized that the criterion suggested by
Muth holds, according to which the predictions of the relevant theory and expec-
tations of agents coincide; even if the functionality itself (i.e. the fact that inflation
targeting works effectively) as it will be seen later, does not follow from the
Muthian definition. Still, huge gaps in our knowledge in economics are becoming
visible. The REH in its rough and pure form does not seem to take production costs
into account (cf. Erdős 1998); according to the classic texts, inflation depends only
on expectations. Monetary policy also relies on the rationality of expectations—
however, some central bank inflation models trace price dynamics back to costs
alone (for one of the most important developments in this research area see
Várpalotai 2003). If, in terms of econometrics, costs alone are enough to explain
inflation (i.e. it is virtually unnecessary to exceed the Keynesian foundations, at
least in inflation theory), what is the role of the (more and more) often cited
expectations within the mechanism? Today, our knowledge is surely imperfect to
provide satisfactory answers.
The role of expectations in price dynamics has been one of the most fundamental
questions of high theory for decades—even if, from time to time, the importance
attributed to it is seen fading away. For example, as early as 1912, Mises pointed out
that price dynamics may occasionally deviate from the base provided by the
quantity of money (during hyperinflation, for example), and that in such cases the
value of money will be determined by expectations (cf. Mises 1981). As for the
progress and improvement of theories that followed, this observation was perceived
to be rather an annoying one, which economists did try to disregard, since what it
implies is the invalidity of the quantity theory (since, as a consequence, it is not the
actual quantity of money that will determine the price level after all). Weird as it
sounds, orthodox monetarists and new classicals did not set aside the rough form of
quantity theory even for a single moment while highlighting the role of expecta-
tions—the scope of expectations was constrained to expectation errors, and the
consequent relation between price levels and quantities of money was, as it were,
conceived of as the outcome of deus ex machina.
The functionality of inflation targeting and its characteristics highlight another
important circumstance. Monetarism (which includes new classical macroeconom-
ics as a radical wing5) established its theory while stipulating a very strong

4
On the effects of rational expectations in a non-equilibrium environment and the efficiency of
anticipated actions of countercyclical economic policies see e.g. Fischer (1977).
5
This statement is not as evident as it is appealing and palatable. Certain authors stress that new
classical macroeconomics may be classified as monetarist (e.g. Mátyás 1984), while others claim
58 2 The Rational Expectations Hypothesis as a Key Element of New Classical. . .

limitation—and this feature significantly confined the relevance of the theory itself.
It appears, therefore, that the monetarist theory can only be applied to describe
macroeconomic systems where the intermediate target of the central bank is to
control the quantity of money. However, to use the quantity of money as an
operative target is a practice no longer paramount6 (it never proved to be very
successful as its applicability has raised problems as a result of a recent break-
through in financial innovation7). The conclusion that can be drawn from all of this
(and is, in fact, often drawn) is that the monetarist theory no longer deserves any
attention.8 However, this is not the case. Modern central bank policy, based on
inflation targeting, is capable of using certain instruments much more efficiently
now (as opposed to how this was done in the past) and, as a result, affecting the
price level with considerable efficiency. Put quite simply, the success of inflation
targeting regimes today stands a good chance of making up for the failure in the
control of the quantity of money. Therefore, monetarism should not be forgotten.
Instead, it should be put on the agenda again and one should consider which
elements of the theory are relevant (in other words, which elements can be made
relevant again) and the relevancy of the central bank’s control of prices in the
context of the REH should be looked into, since this control can be regarded as the
predominance of a tendency resembling the quantity equation. Never should a
theory be regarded as a monolithic entity; it is more like a mosaic whose individual
segments are linked by assumptions and conclusions. This peculiarity has in fact
been recognized by current theories in economics, for in the absence of this
recognition the theory of rational expectations today would not exist without efforts
being made for the entire monetarist theoretical system (whatever is it) to be
rammed down our throats.
In this short chapter, an attempt is made to perform the easiest part of this job,
i.e. the reality behind the underlying assumptions of the REH (the most viable and
most debated component of monetarism) is investigated, given the possibility of a

that orthodox monetarism follows Marshall, while new classicals follow in the footsteps of Walras
(Hoover 1990). Evaluation can be less of a challenge if we believe Milton Friedman, who cited the
Walrasian equilibrium in the context of the natural rate of unemployment (Friedman 1968).
Meanwhile, Frank Hahn does in fact write about the new classicals when describing monetarism
(Hahn 1980).
6
This is distinction is justified. M1 aggregate is a possible intermediate target, while the control of
the monetary base (M0) is an accommodating operative target.
7
On the special aspects and its implementation, along with the instruments involved in this
financial innovation see Vigvári (2004). In this context, one need only think of widely (increas-
ingly so) used e-money, that part of money, used in transactions, which is all but uncontrollable
from the perspective of monetary authorities. Time and again, the development of financial
instruments also underscores the necessity of the redefinition of monetary aggregates (Anderson
and Kavajecz 1993).
8
It must be stressed that the REH and the new classical macroeconomics are not synonyms. REH,
having parted company with new classical macroeconomics, became an independent school of
thought. By any measure, REH is interwoven via its links to monetarism (see the details regarding
the roots of orthodox monetarism and new classical theory).
2.3 The Concepts of Rationality 59

comparison to a directly experienced reality; cf. Weeks 1989. On the other hand,
hidden logical inconsistencies are also highlighted. However, we do not turn a blind
eye to those attractive characteristics of the theory which have led to its eventual
success. First and foremost, our aim here is to paint a coherent picture of what is
known as the rational expectations hypothesis. We do not need to strain ourselves to
see the difficulty of the task at hand as followers of the hypothesis do not form a
homogenous group in terms of the assumptions and conclusions.9 Therefore,
practical difficulties that put a question mark behind the plausibility of the hard
definition of the REH are in the focus of our attention. Considering these problems
will move us on to an alternative (moderated) version of this hard definition which,
stemming from the asymptotic unbiasedness of estimations, accepts the biasedness
of individual expectations and introduces a central institution (i.e. the central bank)
to make the predictions of the relevant (quasi-relevant) model and the expectations
of agents coincide. So, the basic requirement incorporated in the hard definition will
eventually be met, albeit through an alternative and unusual mechanism.

2.3 The Concepts of Rationality

In the literature of the REH, in view not only of new classical macroeconomics but
all other schools that have assumed (a certain degree of) rationality of expectations
in their theory, not a single competent author has worked in the last few decades
without referring to a seminal paper by Muth from 1961 (Muth 1961). Initially, the
idea emerged in a context that differs fundamentally from the framework in which
Robert E. Lucas applied it a decade later. Mention should also be made of the fact
that the rational expectations hypothesis did not emerge in Muth’s paper for the first
time. Jan Tinbergen, in his 1932 paper entitled Ein Problem der Dynamik, specified
a model that resembles Muth’s later work in conceptual terms—with considerable
differences, of course. Tinbergen was aware of the fact that due to the uncertainty
that surrounds the future, agents form expectations (although in an attempt to
simplify the model, he assumed away the difference of expectations). These
expectations are rational (the term vern€ unftig is used in German) if we assume
that they are consistent with economic relations. Moreover, expectations can, in
specific cases, be replaced by the results of an economic theory—and this definition
rhymes well with Muth’s concept and with the concept established in the relevant
economic theory (cf. Keuzenkamp 1991). Recently, the relevancy of the REH has
been questioned on the ground that the future is mostly uncertain, therefore it
cannot be scanned via rational expectations. Thus, imagination has to replace

9
Equally important is the fact that developers and adopters of the REH have relied on the
mathematical apparatus more than every previous school. However, it is also true that their
theoretical inferences and theorems are literally derived conclusions, i.e. they come from the
way equations are specified—to form generalizations is, sometimes, a slow-moving process
characterised by constraints of its own.
60 2 The Rational Expectations Hypothesis as a Key Element of New Classical. . .

reason. It is becoming more and more evident that there is a huge gap between
economic models and the actual mechanisms of the market. Traditionally, these
theoretical models apply static general equilibrium frameworks to predict market
processes, assuming that agents optimize their behaviour while relying on rational
expectations. However, markets are dynamic; moreover, they are characterized by a
process of continuous innovation, self-enhancing, emotional outbursts and uncer-
tainty. In this world of uncertainty, individuals are driven by feelings, emotions,
intuitions and imagination rather than rational expectations or conjectural estima-
tions of future utilities (Bronk 2009).
According to Muth’s oft cited definition, expectations of firms10 (or, to use
accurate wording, the subjective probability distribution of outcomes) will be in
the region specified by the prediction of the theory itself (i.e. the objective proba-
bility distribution of those outcomes) for the same information set. In other words,
rational agents expect an outcome the emergence of which is the most probable—
and this is the criterion by which the rationality of expectations can be assessed.
Muth tries to offer an interpretation of his own theory; the most important element
among the special features highlighted by him is by stressing that a publicly
announced prediction will not exert any influence on the macroeconomic system
under scrutiny, unless the prediction itself is based on insider information, i.e. on an
information set unavailable to market agents.11 Interpretations on Muth highlight
that this particular description of the formation of expectations does not entail the
assumption of perfect foresight by market participants or access to the widest
spectrum of information available—what in fact this is all about is that agents try
to use their information in the most efficient way, while their anticipation is directed
towards the most probable outcomes,12 and they rely greatly on the predictions of
the relevant economic theory while forming their expectations (cf. e.g. Pete 2001).
By the way, the relationship between the assumptions of the REH and those of the
PFH (Perfect Foresight Hypothesis) is often unclarified; sometimes they are
regarded (mistakenly) as synonyms (cf. e.g. Bryant 1983). A key precondition for
a proper interpretation is the need to distinguish between the deterministic and
stochastic models, since the former (and that alone) makes it possible for the REH
to be identified with PFH (Barro and Fischer 1976). In stochastic models, assump-
tions do not determine unambiguously whether or not a certain event will occur—
however, deterministic models are a different kettle of fish: results are evident here

10
It is, perhaps, not an exaggeration to broaden the scope of this statement to describe all market
participants.
11
Rationally formed expectations are not sufficient for the ineffectiveness of discretional mone-
tary policy—to eliminate stickiness is another requirement, while information sets related to
monetary policy and other agents also need to be identical (cf. Dickinson et al. 1982). It will be
argued here that this complementary part of Muth’s thesis, which is ignored by high theory
(i.e. only insider information can modify market expectations) is crucial for the validity of this
theory.
12
It may be confusing, since we shall talk about expectations here as if they were discrete points—
whereas it is obvious that expectations tend to constitute probability distributions. These discrete
points should be regarded as point estimations.
2.3 The Concepts of Rationality 61

(occurrence vs. non-occurrence). This distinction is even clearer if we consider the

fact that a deterministic model delivers a concrete numeric value by way of a
prediction through the substitution of inputs (and under the influence of the
parameters); by contrast, a stochastic model predicts a probability distribution
because of the scope of contingent events. This is exactly what characterizes
statistical and econometrical predictions, since confidence intervals indicating the
probability distribution of a given point estimation can be provided along with
estimations. Reading between the lines, John Weeks’ (1989) analysis seems to
convey the message that refuting the PFH is not that simple. Proponents of this
argument often claim that the volume of information that needs to be gathered and
processed for a perfect foresight far outstrips qualities still possible to be explained
by the utility maximization axiom—according to this line of thinking, forming
perfect foresights would no longer be economical, while its possible emergence
cannot be ruled out entirely. Instead, Weeks voices his criticism on conceptual
grounds making an attempt to deny the implicit presumption according to which
knowing and predicting the future is an act that depends entirely on the quantity of
information processed. It is obvious that it would only be possible to perfectly
foresee the future under deterministic circumstances, where contingencies are ruled
out. Kenneth Arrow himself, by using a suggestive parable, demonstrates that if
contingencies do become part of the game, future outcomes cannot be foreseen with
certainty, though it is clear that possessing and processing more and more infor-
mation can reduce uncertainty (Arrow 1973).
Before moving on, we need to do some more thinking in connection with the
concept of rationality and rational expectations for a while. First of all, mention
must be made of a circumstance highlighted by Muth himself. While reading his
paper, it becomes evident that a rationally formed expectation (or, in other words,
rational formation of expectations) is a cognitive act that can hardly be compacted
in a mathematically specified expectation operator. Whether it be a simple, extrap-
olative (or its purely algebraic derivative, i.e. naive, or regressive) or adaptive
scheme, the process of expectation formation can be described without any problem
by using difference equations. [! expectation operator] Muth, on the other hand,
stresses that rationality (of expectations) is a radically different quality, and this
category does not refer to the formation method through which the numerical values
of expectations are generated (i.e. the expectation methodology) but to the rela-
tionship to processed information (a prediction which is rational in a Muthian sense
can be formed through various methodologies, justified and permitted in view of a
situation). The weakest definition of the REH does not go beyond the claim that
market participants, when forming their expectations, include in their consideration
not only the past values of the given macroeconomic variable but all information
beyond historical data available in connection with the relevant determinants (the
stronger definitions go farther, of course).13 To use an analogy from trade, expec-
tations regarding the future price of Brazilian coffee must deviate from the long-

13
This phrasing should, perhaps, be regarded as the common element in various definitions of
rational expectations.
62 2 The Rational Expectations Hypothesis as a Key Element of New Classical. . .

term trend if a natural disaster were to destroy half of the harvest and that fact
became known to market participants (cf. Pete 2001). However, it must be stressed
that confronting adaptivity and rationality is not necessarily justified, in other
words, there are situations in which following the adaptive scheme is a rational
response. As for the analogy referred to above: if neither a natural disaster occurs
nor any effect capable of changing the trend of a long-term coffee price is seen
emerging, market agents can rely on rational thinking by using all relevant infor-
mation and thus anticipate the continuation of a former trend. In this case, applica-
tion of the adaptive scheme will meet all the requirements in Muth’s strong
definition.
Expectation operator. The relationship between the expected value of an economic
variable (target variable) and the information set used in forming this expectation is
expressed by the expectation operator. In a broad sense, an expectation operator is a
mapping

φ:Ω!E

which creates a link between elements of information set I i 2 Ω applied in expectation

formation and expected values e j 2 E of an economic variable. We can talk about a
mapping as it is not possible that multiple expectations (several elements of the expectation
set) are linked to a certain element of the information set. Elements Ii of information set Ω
are sets themselves, that is,

I i  Ω;

since market agents can, while forming their expectations, exploit several pieces of
information (past numerical values of the target variable, past expectation errors, macro-
economic variables, announcement effects, etc.). So, it is known about the elements of the
information set that Ii :¼ fai ; bi ; ci ; . . .g. The cardinality of subsets Ii of information set Ω
depends on the number of information sources (data sources) applied by agents in the
course of forming expectations. According to the hard definition of the REH, this cardi-
nality can even be infinitely high, since expectations are based on all information available.
Alternative assumptions of expectations suppose a definitely narrower scope of exploited
information stock (Fig. 2.1).
Mapping φ : Ω ! E is surjective, since every element in information set Ω has a
corresponding element in set E, that is, an expected value (but only one element, since
we are talking about a mapping), and there is not any expectation that is not based on some
kind of information. In other words, all elements in set E is related to at least one
information subset Ii. However, this mapping is not injective, since not every element
(subset) Ii 2 Ω is related to different element e j 2 E.14

14
The latter is not particularly hard to realize. If expectations sets consist of, say, rates of inflation
expected for next periods, we can easily find a situation in which the same rate of expected
inflation appertains to different information sets. For instance, an x percent budget deficit and a per
litre fuel price of y make set I1 2 Ω; and a z percent budget deficit (where z < x) and a per litre fuel
price of w make set I2 2 Ω (where w > y). However, agents may link the same rate of expected
inflation to both sets of information.
2.3 The Concepts of Rationality 63

E Ω

e1
I1 I2
e3

I3 I5
e2

e4 I4

Fig. 2.1 Mapping between information set and expectations

The expectation operator as a mapping is a simple function (that can be written in

mathematical expressions with one or more, but finite number of numerical arguments)
sometimes. Since we have defined the connection between these two sets as a mapping,
what follows from this is that in each case the expectation operator is interpreted as a mere
function, in an algebraic sense.15 Although it is true that the quantity of applied information
makes a distinct difference among the possible methods of expectation formation, but
because information and expectation are connected by a mapping, the expectation forma-
tion method can always be regarded as a function (always, in other words, independently of
the cardinality of the exploited information set). It is obvious that the greater the cardinality
of individual subsets I i  Ω is, the more arguments stand in the function. However, it must
be stressed that the forming of expectations is not necessarily a functional and stable
cognitive action (requiring a mathematical expression) but, on certain occasions, more of
a belief. In simple cases, there is an ex ante function (a mathematical formula) that relates
the expected value of the target variable to a finite number of arguments by a stable (simply
put: formulized) operation. Therefore to form expectations is to perform operations on the
information set, i.e. on the arguments where these operations (intake of inputs) are
established by this ex ante function. In other cases, the forming of expectations is not
governed by an ex ante function. A market agent may even process an unlimited volume of
information and relate a given expected numeric value of the target variable to a given
information set (mapping does, therefore, exist), however, the forming process does not
follow a rule (formula, if you like) known by and obvious to the agent. In this case, the
forming of expectations can be regarded as a belief.16 As in this case relation is also driven
by a mapping, the ex post expectation function can be identified (at least theoretically, i.e. if
we know the information set on which the agent formed his expectations). According to the
hard definition of the REH, it is not possible to formulize such an ex ante stabile relation
(function) and primarily not for the infinitely high volume (cardinality) of information
exploited in the forming process but because the characteristics of information continu-
ously affect the forming method. In case of a certain information set, a simple extension of

15
Mathematicians often label functions simply as mappings or operators (see e.g. Rozgonyi and
Toledo 2008). It has to be added that functions do not have to be mappings between sets consisting
only of numerical elements. In terms of the identification of mappings (functions), the one-to-one
character and not the nature of elements has a crucial importance.
16
However, no consideration precludes the case in which an ill-informed agent forms his expec-
tations as beliefs rather than on the basis of an ex ante formula.
64 2 The Rational Expectations Hypothesis as a Key Element of New Classical. . .

the previous trend may be appropriate, but if information changes, another method may be
needed.
In a narrower sense, the expectation operator is a function with only a few arguments
among the inputs of which only the lagged values of the target variable (xt) and, perhaps,
past expectation errors are present. The simplest extrapolative model can be written in the
form of

xte ¼ xt1 þ αðxt1  xt2 Þ;

where xet denotes the expected value of xt formed in period t  1. It has to be noted that if
α > 0 then xet is the extrapolation of xt1 reflecting the most recent change in the variable
and following the direction of this change. If α ¼ 0, it is the plainest naive scheme, that is,
agents anticipate a simple continuation of present. In the third case, if α < 0, the regressive
scheme starts predicting the former value of the given variable.17 It is the substrate of the
adaptive model that the expectation error (i.e. the difference between the expected and the
actual values) calculated on the basis of a former (computed usually for period t  1 )
expectation is added to the formation of period t expectations as a correction factor. In this
case, our operator can be defined as


xte ¼ xt1
e
þ β xt1  xt1
e

in algebraic terms, where β > 0:

Expectation operators also play an important role in econometric modelling as well. By
hypothesizing adaptive expectations, you can construct an equation on current and lagged
data of income and consumption expenditures which leads you to a numerical estimation of
marginal propensity to consume out of adaptively expected income and, hence, to the
possibility of testing your hypothesis (Ramanathan 2002). Maddala (2001) offers examples
for the treating of naive expectations in econometric terms. However, it is common in all
these cases that there is no observed data available for expected values (at best, one can
identify the expected values from de facto observations just as it is done by Maddala). If,
however, you have access to explicit expectation data,18 and you are familiar with the

17
Visco (1984) offers an excellent summary about various expectation operators and the possible
applications.
18
In Hungary, following the methodology established in North America and Western Europe after
World War II, the Economic Research Company (GKI) prepares surveys on a monthly basis
through which market participants’ inflation expectations can be revealed. GKI surveys five
groups of market participants (industry, trade, building and construction, services and households)
so that expectations regarding relevant future market outcomes and opinions about recent devel-
opments can be reviewed. GKI has been providing consumer surveys since March 1993, accom-
panied by business surveys from January 1996 onwards. The business sector is represented in the
survey by being divided into various branches: agents from industry and trade are surveyed on a
monthly, while market participants from building and construction plus services were initially
asked on a quarterly basis. From January 2002, all the business sector has been examined every
month. In the course of the consumer survey, 1,500 individuals are visited by surveyors (the
number of those interviewed was 1,000 before May 2001). All in all, the GKI, the Institute of
Marketing and Media of the Corvinus University in Budapest also make consumer surveys on a
quarterly basis, following the methodology of the University of Michigan (Index of Consumer
Sentiment—ICS) the main output of which is “Michigan Consumer Sentiment Index”. Sampling is
2.3 The Concepts of Rationality 65

information set on the basis of which these expectations were formed, you can specify the
ex post expectation function that governs the formation process.19

But what about the weak definition of REH? In its briefest form, it states only
that rational agents form their expectations on a relevant (economic) variable
through efficiently exploiting all public information on factors which they think
are capable of affecting the future path of a specific variable (Horváth 2000).
Apparently, this definition reveals next to nothing in terms of the expectation itself,

done in two stages (first: settlements and, second: individuals are picked randomly); the sample
should be representative in terms of gender, age, qualification and types of settlement (Vadas
2001). The ratio of successful interviews is, generally, over 90 %. According to econometric
investigations, consumer confidence index predicts with accuracy the dynamics in consumption
expenditures (consumer confidence index holds additional information about the overall condition
of households and the uncertainties of future income stream), therefore the Central Bank of
Hungary relies heavily on it for information processing and forecast purposes (Jakab and Vadas
2001).
In the course of business surveys, 5–6,000 enterprises (having legal entity and with 20 emp-
loyees at a minimum) are asked on a monthly basis via postal questionnaires. Restricting the
sample to companies over 20 employees has proved to be a technically and methodologically
competent decision. Thanks to this, a greater stability of samples can be guaranteed, while data
consistency and reliability continue to be a high priority rather than being something of secondary
importance, given that experience has shown that the smaller a firm, the more unreliable
(systematically biased) the data provided by it. Small enterprises (with a staff fewer than
20 employees) are units usually relying on family and acquaintance relationships, concentrating
on local markets and geared to provide maintenance for the family—therefore they can hardly be
separated from the household which provides the infrastructural background and capital sources.
The attitude and the goals of these enterprises may radically differ from those of larger businesses
(Toth and Vincze 1999; Toth 2002). The sample comprises 1,500 processing industry firms. The
sampling method is quite unique: the sample available from the previous year is updated every
July so that 300 new (that is, previously not involved) firms enter while 300 firms are removed
from the sample. This routine biases towards enterprises being already in the sample, which
restricts the possibility of deletion—while the sample functions as a quasi-panel. The advantages
of panel studies over longitudinal surveys are quite obvious as in this way researchers can gain a
better understanding of the changes that affect individual units or groups of those units
(cf. Babbie 1989). The number of responses (completed questionnaires) from the entire sample
(i.e. not only from processing industry firms) is around 1,000–1,200; propensity to respond is
slightly under the EU average, while it is not particularly low compared with former socialist
countries. Responses are representative, both territorially and sectorally, i.e. responding firms and
the entire sample do not differ substantially. By contrast, there are significant differences between
responding firms and the sample in terms of business size as experience suggests that propensity
to respond is significantly lower for small enterprises; according to GKI reviews, this is not
particularly problematic, since in Hungary medium-size and large companies do, in many ways,
play a key role. Active, responding businesses reveal homogeneity to a large extent (both in terms
of sectors and individual companies), i.e. the circle of enterprises that fill out questionnaires in
each month can be regarded as rather unchanging, thereby helping data processing (T oth 2002).
19
The distinction between ex ante and ex post expectation functions suggested here bears strong
resemblance to the potential roles of monetary policy Taylor-rules. A monetary policy setting the
key policy rate for the next period may explicitly follow a specific Taylor-rule (ex ante role). Even
if the central bank does not follow such a rule, a Taylor-function as a regression equation still
describes well the interest rate decisions of the monetary authority subsequently (ex post role).
This will be a recurrent theme in Chap. 6.
66 2 The Rational Expectations Hypothesis as a Key Element of New Classical. . .

except a little bit about the formation process—and, according to this definition,
expectations regarded as rational will be different from those included in the hard
concept, where reference to probability distribution is tantamount to a significant,
additional knowledge. As for the coffee price analogy mentioned earlier on, a
market player who optimizes his behaviour through comparing costs and benefits,
gathers and processes information up to a point when its costs exceed the benefits
hoped for (or avoidable losses). This weak definition claims that an investor can, by
thinking rationally, expect the coffee price to keep in line with the former trend
even after the occurrence of a natural disaster affecting the plantations, if the
expenses of collecting and processing information required for the changing of
expectations are higher than his (potential) loss resulting from an incorrect expec-
tation. However, it is clear that this critical point is different for every investor, as
certain investors in the same situation risk losing low amounts due to incorrect
estimates, while others might be exposed to a considerably higher loss, depending
on the actual amount of investment. And, if this is the case, the convergence of
individual expectations will be out of the question, given the discrepancy in the
expectations due to the information process having discontinued in various phases
(moreover, these expectations may be systematically biased). Therefore, according
to the weak definition, a simple trend extension by a small investor may be regarded
as being rational. In this case, the only stipulation regarding rationality is that the
agent should keep processing information up to the optimal point and expectations
formed accordingly will be labelled as rational.
Experimental economics. Experimental economics has discovered the laboratory for
economic theories as well—the laboratory which was already known to and used by
sociology and social psychology (cf. Babbie 1989; Bungard 1997) or marketing science
(e.g. Malhorta and Simon 2009). Following that, economics was no longer in want of
experiments. The first attempts were made by Edward Chamberlain (Caginalp et al. 2003),
and experimental economics was eventually awarded the Nobel price thanks to Vernon
L. Smith, Chamberlain’s follower.
At the outset, experimental economics represented a definite methodological
programme that can be interpreted as one surpassing the traditional schemes of economics
(KVA 2002). According to experimental economics, mainstream economics abounds in
theses that are no more than assertions never confirmed. It is, perhaps, nothing more than a
belief that macroeconomic systems can be fine-tuned in accordance with the Keynesian
suggestions, or that the problems caused by monopolies can be solved by regulatory
measures, or, if exposed to certain externalities, the market is doomed to failure. One
objective of experimental economics is, therefore, to test the fundamental theses of
economics (Smith 1982). Further efforts were directed towards delivering data upon
which new theories can be based. Through experiments, certain behavioural tendencies
can be identified upon which to build a theory at some later point, capable of interpreting
these tendencies in clear and explicit terms. (Smith 1994).20

20
This is as trivial in meaning and importance as it is objectionable. At one point, Smith (1994)
writes: “Well-formulated theories in most sciences tend to be preceded by much observation,
which in turn stimulates curiosity as to what accounts for the documented regularities.” This is the
very essence of an induction-based theorizing routine. Common theoretical-scientific thinking
finds nothing exceptionable in induction—on the contrary, it is regarded as the fundamental
2.3 The Concepts of Rationality 67

However, to do nothing other than detect and describe the laboratory behaviour of
human beings was not the ultimate goal, but to identify the building blocks of which actual
market mechanisms and macroeconomic systems consist. Neither astronomy, nor meteo-
rology, nor theoretical economics is able to conduct experiments with the objects (planets,
stars, weather fronts or national economies) that they observe. However, the results of some
laboratory experiments can be exploited directly in studying astronomical and weather-
related data and in interpreting observations, since the same natural laws are valid for the
whole physical universe. According to experimental economics, the same is true for the
situation when socio-economic systems are investigated as the fundamental elements (self-
interest, motivations, risk-aversion) of behavioural effects that can be identified in real-life
situations (i.e. reality lived through via direct experience) emerge and also work in
laboratory experiments (Smith 1976; 1989).21 [! phenomenology]
Most important among the findings of experimental economics was, perhaps, the
observation and description of unconscious optimization strategies. While theoretical
economics assumes humans to be rational, psychology highlights their irrational character
(Smith 2003). By contrast, experimental economics, rather than rejecting the concept of
rational human behaviour accepted by mainstream economics, is set to fine tune it and also
to provide more of an in-depth view of it with the aim of strengthening the consistency
between mainstream theory and observations. Mainstream economics postulates the indi-
vidual’s rationality as the result of a conscious, cognitive process. If optimization were an
unconscious process, how would it take shape? In the wake of a miracle or casual
circumstances? This is exactly the view which experimental economics seems to abandon,
while stressing the idea that the concept of market equilibrium cannot be separated from the
learning processes that are essential for its creation. (Smith 1991). Despite the discrepancy
between experimental economics and neoclassical orthodoxy seems to be serious, there is
still no reason to talk about a conflict. Experimental results have often confirmed that a
REH-type (or a near to REH-type) equilibrium may be established not only in situations
depicted by abstract models but also in laboratory conditions. We might also envisage a set
of rational expectations, although this rationality is not the outcome of innate skills, for its
emergence takes time and experience, i.e. the observation of market processes and the
behaviour of other agents22 (cf. Smith et al. 1988; Porter and Smith 2003).
At the same time, these experimental results have efficiently supported the mainstream
theory. It was realized during these experiments (and, particularly, through post-experiment
interviews) that agents achieve the utility and profit maximizing equilibrium of mainstream
economics without being aware of this. Although their perception of these experiments,
designed as models of real-life economic settings, identified these situations as chaotic
messes, an optimal (or near-optimal) state of equilibrium characterizing perfectly free
competitive (i.e. theoretical) markets was created without the players being aware of this.

method applied in the process of theorizing. For instance, Popper, on grounds of Hume’s
philosophy, refuted induction, since, accordingly, there is no pure observation at all which
might be independent of any theoretical grounds (as opposed to Vernon Smith’s interpretation),
therefore it cannot serve as a starting point for theorizing. For him, therefore, observations are
selective and are governed by theories (cf. Thornton 2009).
21
It is interesting to realize that, based on this, Friedman’s assertion (Friedman 1953) that social
and natural sciences can be hardly distinguished on methodological grounds holds after all. The
possibility of performing controlled experiments in economics does not imply that the border-line
between social and natural sciences will disappear automatically.
22
Accordingly, bubbles should be always expected to emerge under life-world conditions since,
although practised players (experienced market participants) are fully aware of the fundamental
laws of market, there are always agents who, driven by the lack of experience, actively contribute
to the emergence of bubbles (Caginalp et al. 2003).
68 2 The Rational Expectations Hypothesis as a Key Element of New Classical. . .

(Smith 1991). Market rationality does not require perfect and complete individual ratio-
nality. The only plausible interpretation of these findings is the pervasive presence of the
invisible hand. Market agents can realize socially efficient outcomes with no intention to do
so (Smith 1994). The pursuit of a state of equilibrium is an unconscious, cognitive process,
and the abstraction-based, pure economic theory performed quite well in outcome pre-
dictions. [! paradigm]

It is hard to interpret the optimal quantity of information obtained which, as we

have seen, may also necessitate the availability of estimates regarding costs
incurred through collecting excess information and the benefits derived from
obtaining that information. According to a different approach to reach optimum
levels, information available should be regarded as optimal as long as it is capable
of fundamentally affecting a decision. The only problem is that this optimum can be
evaluated only ex post. In this case, information—which did not change an earlier
decision that continued to be exposed to (repeated) modifications as new pieces of
information were being processed—is considered superfluous. However, in order
for this to become obvious, it is necessary to obtain information which might
subsequently become redundant (in other words, we do not know ex ante that a
particular unit of information will eventually be useless). This definition is unac-
ceptable, since presupposing the ðn þ 1Þth piece of information to keep our decision
intact might be erroneous, if our expectation is based purely on the fact that the nth
item did not force us to modify our expectation. That presupposition implies the
view that relevant items are always arranged in an order of importance and if a
processed item turns out to be futile (in the sense that it has failed to trigger the
modification of our decisions), we can draw the inference that no more relevant
information will be obtained. Moreover, optimum can be described as a subjective
optimum, in which case an agent, while forming his expectations, will be led to
believe that he has already collected and processed all necessary information. In
order to avoid all these dilemmas, the sociological theory of rational choice uses a
simple temporal concept of optimization, i.e. it regards the gathering of information
a course of action to pursue only as long as this process does not endanger decision-
making (cf. Elster 1989). However, the rule here is confined to stating that it is
pointless to search for new information beyond the point at which a decision might
have been taken..
It has to be noted that the weak definition of REH refers to the relation between a
belief and the grounds on which it is held (we will revisit this problem shortly for
further details) and, therefore, this concept is in close relationship with definitions
used in the literature of sociological theory. One should not confuse the relationship
between a belief and its grounds with truth, which can be interpreted in the context
of the relationship between a belief and what the belief is about (cf. Elster 1989). If
we follow this line of reasoning, Othello is seen entertaining a rational belief when
he believes that Desdemona betrayed him, since this inference seems to be logical
to him on the basis of the information he possesses. Indeed, the weak definition of
REH does not venture beyond this. In other words, rationality does not require
beliefs to be true (or punctual). Put differently, according to the weak definition,
2.3 The Concepts of Rationality 69

rational expectations are not required to coincide with subsequent de facto out-
comes or with the expectations of their previously known probability distribution.
However, the hard definition of the REH goes considerably further, also linking
the relationship between belief and its object to the concept of rationality; more-
over, it seems to focus exclusively on it, that is, correctness is involved in the
criteria while truth is still not a requirement, since belief is not aimed at a current,
verifiable fact. If our aim here is to apply a concept in the context of the facts that
are verifiable at the present time, one which is as rigorous as the hard definition of
REH, we should require rational beliefs to be true: Othello’s suspicion cannot be
regarded as rational as Desdemona did in fact remain faithful. At the same time, it
should not be forgotten that the REH deals with current estimations of future
outcomes, and if truth were a prerequisite in this temporal relation, it would be
equal to insisting on the PFH—however, a perfect foresight is not possible under
stochastic conditions. The present knowledge of the probability distribution of
future outcomes is a manifestation of most of the existing knowledge regarding
the future—therefore the hard definition of the REH requires an ability to look into
the future with utmost precision which, of course, cannot go further than requiring
expectations to follow the objective probability distribution of future outcomes.23 It
is the stochastic nature of socio-economic processes alone that excludes truth from
our list of criteria.
We can say that the strong definition of the REH offers a criterion on the basis of
which beliefs can be classified as rational or non-rational without a subjective
value judgement. Such a labelling would be considerably more problematic in the
context of the weak definition, since an external judgement regarding the rationality
of a belief implies knowledge about knowledge, i.e. the information set, based on
which a particular agent made his decision, needs to be known (and the one relying
on which he might have made that decision). Mention should also be made of the
confusion caused by the augmenting part in Muth’s definition. The clause on the
shared information set is a visible shift towards the weak definition, since if a
cluster of agents has access to less information than the inputs of the relevant
model, their expectations formed under these circumstances may be still regarded
as rational according to Muth, although they will be biased. We can, therefore, say
that both the hard and the weak definition can be traced back to Muth’s definition,
while highlighting different aspects of it. Applying Muth’s (complete) definition
would get us into a situation too comfortable to handle (but confusing), since even if
the requirements of the hard definition are not met, the weak definition may still be

23
It seems that the problem caused by temporality is not obvious for sociologists, i.e. they may
question the rationality of a decision or belief on future outcomes if it turns out to be false
posteriorly (!) (cf. Farkas 2006). There is a fable about a person looking for a new job, who
accepts an opportunity, thinking that this job is ideal for him in terms of his preferences. He is
assumed to exploit all relevant information available before making his decision. However, he
becomes disappointed later, since he has less leisure time or earns less than he expected. Some
authors think that the rationality of his decision (his expectation) can be doubted—without
realizing that what they do is require foresight (i.e. pre-knowing) into the future while even they
do not have such a knowledge (since this doubting occurs only posteriorly).
70 2 The Rational Expectations Hypothesis as a Key Element of New Classical. . .

put in place (with all its inconsistencies). In this case, the most serious problem
would be caused by the fact that the weak definition does not put forth a measure
based on which one can regard an expectation as rational. We have seen a
significant difference between requirements implied by the weak and the hard
definition, merging these two concept should, therefore, not be advocated. It is
not surprising that the new classicals subsequently focused exclusively on the hard
definition (accordingly, the REH will from now on be identified with its hard
definition).
It is worth considering what the criterion of rationality embedded in the weak
definition might be. One can say that an expectation is not rational if a particular
agent stops gathering and processing information before achieving an optimum
level—in this case, it is pointless to investigate the relationship between a belief and
its grounds, since as the information available is insufficient. An agent, confident of
success, who is known to have invested huge sums of money in the Brazilian coffee
business, can hardly be labelled as rational if the information gathered on future
coffee prices was obtained from, say, the evening news or from next-door neigh-
bours (except if his neighbour runs a coffee plantation in Brazil), even if his
expectations are in synch with the information set upon which those expectations
rely. In this case, the question mark behind rationality does not stem from the
assumption of irrationality (although the everyday meaning of the term “rational-
ity” undeniably suggests this) but from the fact that the agent in question had no
access to the optimal amount of information required. If the weak definition were
limited to the scrutiny of the relation between a belief and its grounds, the agent
who expects the extension of the previous trend on the basis of an effortless
performance aimed to gather information, and is unaware of the natural disaster
and the damages caused by it (consider the case of the small investor referred to
above) would be thought of as someone behaving in a rational manner. However,
according to the weak definition, there is a different option whereby the notion of
rationality might be denied. Even if the market agent under scrutiny did obtain an
optimal amount of information (whatever optimal means), he could still form an
expectation based on that very information, which would contradict his previous
knowledge on the matter (in other words, his belief is not in accordance with the
information processed). Cast your mind back to the classic example: the attitude of
the agent who expects coffee prices to be unaffected after the harvest has been
destroyed is not rational, either. Thus the weak definition specifies two criteria
without clarifying the connection between them. Moreover, it should also be noted
that the assertion (or assumption) formulated in line with the weak definition that
agents are rational, is not much use from an analytical perspective. Even if the
assertion itself is correct, expectations may well be infinitely heterogeneous, and
the analytical apparatus along with the economic policy conclusions of new clas-
sical macroeconomics could hardly have been based on this concept.24 Yet the most

24
However, there were attempts in the literature to link the implicit acceptance of the weak
definition to the conclusions of the hard definition. However, these are not logically consistent
efforts (cf. e.g. Shaw 1984).
2.3 The Concepts of Rationality 71

serious problem is caused by the far too cautious wording of the weak definition
and, as a result, its observations do not venture beyond what has been well-known
for generations of economists, their view being grounded in common sense
(i.e. agents gather information in order to make decisions and form their expecta-
tions on the basis of that information).
Although the hard definition avoids all these problems, it creates additional
difficulties. The only prerequisite of rationality here is for agents to form their
expectations in line with the probability distribution. With a definition like this, to
establish a view poses no difficulty: an agent whose expectations coincide with the
anticipated outcomes of future processes will form rational expectations. The
problem of fudging the issue by the ambiguous interpretations of optimum does
not need to be addressed, nor is ‘knowledge about knowledge’ a must anymore—
although, as it turns out, the smokescreen created by the non-clarified background
regarding expectations is one of the most serious problems associated with the hard
definition.
When dealing with the weak definition, the lack of hard definition-type, unam-
biguous standards raises the problem of subjective judgement. It is obvious that
rationality is either questioned or recognized by an external observer (for instance, a
specialist examining the issue from the vantage point of science). Even if it is
assumed that this individual is familiar with the information set possessed by the
agent under observation, and if it is also assumed that this set is optimal in quantity,
a disagreement on the inference drawn on this basis is still possible between the
spectator and the agent observed. The connection between a belief and its grounds
is not based on rock-solid stability. Once again, the coffee plantation analogy
comes in handy: even if the investor is informed on frost damages and even if he
expects an upturn—to what extent should price fluctuations be considered rational
by that agent? If the investor expects a 25 % rise, while the observer predicts a 50 %
increase (or vice versa), this will inevitably cast a doubt on the rationality of
expectations. Accordingly, the weak definition is vulnerable to queries (apart
from its weaknesses mentioned already) for leaving the relationship between a
belief and its grounds unclarified (while passing judgement on it). At this point, the
hard definition makes reference to the relevant economic theory that delivers a
well-defined output—therefore, standards in this case are unambiguous. The weak
definition, on the other hand, stemming from an ill-defined set of information (since
every agent may have obtained dissimilar sets of information) and in the absence of
an unambiguous method for the formation of expectations (i.e. reference to the
relevant economic theory and macroeconomic model), does in fact make the claim
that behavioural decisions can be reached (to use a simile by Berlin 1958) with the
help of a slide rule. The attitude of a spectator basking in the scholar’s role and
assuming the superiority of his observations over the views taken by the agent under
scrutiny, may be criticized from various aspects. Moreover, the links between this
debate and the one on paternalism are hardly less than direct (cf. Mill 1859).
72 2 The Rational Expectations Hypothesis as a Key Element of New Classical. . .

2.4 Rational Expectations and Unbiasedness

After this brief digression, let us return to the hard definition given by Muth (1961).
Simple phrasing can raise doubts. If the distribution functions (and the probability
density functions derived from them) of the predictions provided by the relevant
model and by the economic participants are identical, it follows that the predictions
offered by market agents are unbiased (since the expected values are identical), and
even the variance of the estimations (i.e. the average of the squared differences
from the expected value) will be identical. Therefore, the hard definition of the REH
would imply that market participants anticipate the same outcome as the one
predicted by the relevant theory. However, this mechanism cannot solidify into a
perfect foresight as errors in expectation cannot be excluded. In other words, the
claim that expectations of agents are all identical will eventually prove false (and
we have to admit that a perfect foresight, which rules out the possibility of errors
per definitionem, would imply that expectations are identical). Our suspicion that
Muth believed the predictions of the theory and agents to have identical distribu-
tions is supported by the fact that Muth precluded the possibility of better estimates
based on the theory. Therefore, there is a fundamental difference between the
output of the adaptive model (and other models referred to above), which relies
exclusively on past information, and the rational scheme, since a systematic bias
from actual outcomes (and from the predictions of the relevant model) may
naturally occur in the former models—while in the case of rationality this bias
would, in theory, be ruled out. However, as we shall see, it is precisely that
unbiasedness which is the most problematic element of the REH, the assumption
of which can only be confirmed amid grave concerns. The lenient approach
manifested in the reasoning that the underlying intention here is to efficiently
exploit information and that utility maximization is the only explanatory variable
in the process of gathering and handling information (cf. Kantor 1979). This can
explain the motives at best. Although there is no denying that the REH highlights
the importance of exploiting information, the methodology of the process itself
remains unrevealed.
Some additional remarks need to be made with respect to the details outlined in
the previous paragraph. Although Muth seemingly precluded the possibility that the
theory can lead to better (sic!) predictions than the expectations of firms if ratio-
nality is assumed, he drops a hint at some point that the predictions of individual
firms (agents) may be subject to greater errors than the theory itself.25 Undeniably,
this is a cautiously worded assertion, just like the one where Muth stops short of
making the unambiguous claim that subjective and objective probability

25
It implies that the predictions of the relevant theory are not true either, i.e. neither the relevant
model nor the aggregated expectations meet the requirements of the PFH. Some definitions of the
REH are more daring, making the claim that aggregated expectations are true (Begg 1982).
2.4 Rational Expectations and Unbiasedness 73

distributions are identical. All he says is that these two distributions tend to be
identical.26 This apparent contradiction will be resolved, since Muth notes that,
when expectations are rational, agents’ aggregated (i.e. not individual) expecta-
tions will coincide with the output of the theory (and, therefore, probably with the
actual outcomes). And although it is true that there are random errors which come
between the individual predictions of agents and data subsequently obtained, and
that the expected value of errors is zero (to ensure that aggregated individual
expectations coincide with the predictions of the theory), complete with orthogo-
nality, a criterion to be tackled later, there must be (according to Muth) a divergence
in the absolute values of errors, governed by unrevealed variables. Therefore, one
should not remain indifferent (not even on the face of it) to the problem of who
those expectations are formed by. Aggregated unbiasedness is an existing assump-
tion, anyway; systematic errors are thus precluded.27 Individual predictions are
unbiased at the aggregate level, while the relevant model is capable of delivering
unbiased estimations. However, it is quite obvious that hardly anyone is familiar
with the relevant model (not in its entirety, anyway) given the fact that the view put
forward by Muth on this issue is that there are considerable cross-sectional differ-
ences in opinion (first conditional is absent from the wording) from which one may
even infer the biasedness of individual expectations, because, as it will be seen,
there is a strong likelihood that predictions based on incomplete information will
lead to biasedness.28 Muth mentions that the averages of expectations (i.e. at an
aggregate level) are as accurate (sic!) as the outputs of a complex equation system,
implying that the model is not used in the estimation process. It follows from this
that these expectations are generated through a different method. Moreover, Muth
also makes the point that his theory falls short of declaring any resemblance
between agents’ estimation methodology and complex econometric models,
although it does say that the way in which expectations are formed depends on
the structure of the relevant model describing an economy. These two kinds of
predictions (i.e. those of market agents and of the relevant theory) are, therefore,
not identical but entirely similar. This has serious consequences, since apart from
having to reveal the structure of the model (and how knowledge relating to that
model can be acquired), an explanation is also required to clarify how market

26
“[. . .] expectations of firms (or, more generally, the subjective probability distribution of out-
comes) tend to be distributed, for the same information set, about the prediction of the theory
(or the »objective« probability distribution of outcomes)” (Muth 1961) (italics added). The
question is still, first: what happens if exploited information sets largely differ and, second:
whether unbiasedness can still be maintained under such circumstances.
27
Under stochastic conditions, unbiasedness means that the expected values of parameter esti-
mates are the estimated theoretical parameters and random errors prevent theoretical and empirical
parameters from coinciding.
28
OLS-estimations always guarantee the expected value of errors to be zero—however, one
should not infer the unbiasedness of the model or its predictions. Moreover, it is also true that
the less information is exploited during the specification of the model (that is, the more relevant
variable is precluded), the higher the sum of squared residuals will be. One can hardly achieve
unbiased estimations with relevant variables being omitted.
74 2 The Rational Expectations Hypothesis as a Key Element of New Classical. . .

agents can acquire knowledge that in actual fact is not identical with familiarity
with the relevant model, only entirely similar to it (at least in terms of its outputs).29
In fact, the situation would be less complicated if it could be assumed that
economic agents are experienced in econometrics and economic theory. In this
case, if it is true that all these agents and the relevant economic theory possess
information sets (including the methodological knowledge that is necessary to
specify the model) where each set is equally valued (with random differences in
their contents), individual estimations are all unbiased estimations of actual pro-
cesses and the relevant model will be in the centre of the probability distributions of
parameter estimations. However, theoretical parameters are not known—only the
applied functional forms (agents would, in this way, both know and apply the
relevant model, even if the job of filling it with inputs would be a task left to
individual agents). Thus Muth’s requirement according to which expectation aver-
ages are true (unbiased) would be met. However, this explanation is not consistent
with Muth’s concepts, since, in his view, the methodology applied by the agents is
not based on econometric modelling. This view could explain why agents make
major forecast errors, but on the level of individual estimations it would result in
bias due to the primitive nature of expectation formation methods—however, it
would become clear why errors show cross-sectional differences (since it follows,
among other things, from the details above that better-informed agents can make
better estimations). However, in this case, no reasonable explanation can address
the issue why the average of biased individual estimations should be an unbiased
estimation of actual processes.
These two possible explanations (agents are well-informed professional econo-
metricians vs. agents have naive-primitive expectations) are differently consistent
with the definition of Muth. The first option is consistent in terms of its conse-
quences: individual estimations are unbiased. The second option is consistent in
terms of its assumptions: the method associated with agents is not based on complex
econometric methods. It seems that these two points in Muth’s definition can be
hardly held simultaneously. Moreover, the third option, in which Muth might have
suggested the adoption of the outputs from the relevant model by agents, can be
ruled out, as in this case there would be no cross-sectional difference in expecta-
tions. Thus the as-if-argument itself pales into insignificance, since if it were true
that agent expectations are being formed similarly to the outputs of the relevant

29
Muth also makes a riddle by claiming that “Averages of expectations in an industry are more
accurate than naive models and as accurate as elaborate equation systems” [. . .]. Now, we are
faced with the fact that predictions of market participants are not made on mathematical grounds
(at least not by using complex methods). It is exactly that reasoning bluff which was objected by
Benjamin Friedman and John Weeks to new classicals (and to which the obscure hint on the
approximating distribution functions mentioned above bears resemblance): expectations are being
formed as if agents know the relevant model (B. Friedman 1979; Weeks 1989). Remember the
F-twist and the parable of rational and mathematically optimizing leaves. Muth simply does not
recognize that his concept of market agents could be useful as an ideal type even without this bluff.
2.4 Rational Expectations and Unbiasedness 75

model, we would be left in the dark as to where the cross-sectional differences of

opinion mentioned by Muth are derived from.
It becomes clear that new classical macroeconomics, by using the REH, comes
up with arguments based on the notion of assuming the conclusion (that is, petitio
principii). Unbiasedness is an important criterion because new classicals deny the
efficiency of a systematic countercyclical monetary and fiscal policy on these
grounds—if there is no bias, there is no room for stimuli via monetary policy,
either. And new classicals hold the view that the converse is also true: if the control
of output is not feasible with the help of monetary policy, from this we may infer the
unbiasedness of estimations (or, to be more exact, we should infer it, since the REH
postulates per definitionem unbiasedness).30 Our most important question is, there-
fore, whether unbiased expectations are an over-simplifying assumption or a
plausible presupposition by the REH, since if it were understood and accepted
that to adhere to this assumption is not a realistic option, this finding would
considerably impair the relevancy of the REH and, accordingly, the reliability of
economic policy conclusions put forth by new classicals.
We cannot turn a blind eye to the fact that there are two kinds of expectations and
estimations to deal with during our scrutiny of the REH as we need to make a
distinction between the
1. Predictions of the relevant model and
2. Expectations of individual agents.
At the same time, we shouldn’t overlook the fact that the relevant model is in fact
unwanted, unless if it is used for the purpose of spearheading the as-if-expectations
of agents. The intention to deceive by a monetary policy is directed towards the
actors: the point here is that their expectations are unbiased—while the unbiased-
ness of the relevant model (if there is such a model) is, in actual fact, an insignif-
icant circumstance. Therefore, the problem of what are the prerequisites of
unbiased estimates provided by the agents and those provided by the relevant
model itself, needs to be viewed in a comprehensive manner. As a starting point,
market agents will be assumed to be professional entities producing econometric
models, otherwise unbiased individual estimations could hardly be assumed (more-
over, to assume that agents have this knowledge is acceptable in a pure theory—the
usefulness of these assumptions in a theory aimed at producing a realistic descrip-
tion of the life-world is a different story). If the conclusion drawn from this is that

30
Here, some attention should be paid to Weeks’ reasoning, according to which the REH is also in
error by believing that reality is governed by deterministic laws. Accepting it is really problematic.
If there exists full knowledge on (an) economy (as the REH states), future outcomes occurring
deterministically could be forecast truly (and not only unbiasedly). In this case, the REH and the
PFH would be equivalent even at the level of agents—but this is not proclaimed by the REH.
Therefore, on this account, it is not determinism as far as explicit assumptions considered at least.
It seems that the accusation of determinism is trumped-up for rather a prosy reason: it is revealed
from Muth’s phrasing that not even theory is able to provide true but only unbiased predictions.
However, it is a different problem that the PFH is very much in need, say, for operations of the
labour market—labour market as depicted by new classicals (for further details on this problem see
Chap. 3).
76 2 The Rational Expectations Hypothesis as a Key Element of New Classical. . .

agents are capable of producing unbiased estimates under certain conditions, our
criticism should be directed at other areas, i.e. the conditions themselves, by
making the point that, say, an individual who runs the local food store can hardly
be expected to have a particularly high-level expertise in mathematics. The exis-
tence of the relevant model, the possibility of its successful specification and the
unbiasedness of the estimations of agents do not presuppose each other, and in
terms of the economic policy consequences, unbiasedness is the only relevant
factor. And if the necessary conditions in the context of knowledge and information
are not met, which means that actors cannot be expected to provide unbiased
estimates, unbiasedness can only be ensured through the special as-if-assumption
of the REH (which in turn would obviously need to be justified).

2.5 The Orthogonality Assumption of the REH

So far, we have considered the conceptual difficulties caused by the REH presup-
position that expectations are unbiased. It is quite obvious that we are forced to
circumvent the issue if we want to follow the reasoning of the theory: by using
non-complex methods, individual agents are not likely to end up having unbiased
estimates. Agents would need professional modelling skills to achieve this, but
Muth precluded this option in advance—his rules regarding modelling techniques
will exclude everything except biased estimations. Unbiasedness is nothing but a
presupposition in the narrowest sense of the word.
However, the REH also assumes orthogonality along with unbiasedness. Unbi-
asedness means that the expected values of residuals defined as the differences
between data (as probability variables) and predictions are zero. By contrast,
orthogonality means that expectation errors do not correlate with the information
set available. Therefore, according to orthogonality, the ex post expectation errors
(i.e. those calculated on factual data) cannot be forecast ex ante on the information
set available and exploited when expectations are formed (Gerrard 1994). On
second thoughts, this turns out to be a necessity, since if it were possible to forecast
expectation errors in advance, expected errors themselves would become parts of
the information set available (ex ante), which would thus be capable of modifying
expectations that are being formed—and only random errors might occur (which
would fail to take us to the fairly powerful assumption manifested in the PFH).
Therefore, according to the REH, the rationality of expectations is confined to the
precluding of systematic prediction-expectation errors while, as has already been
mentioned, random (i.e. unpredictable) differences (random, in the interpretation of
the REH31) may still occur. Therefore, the requirement of orthogonality can be
described in the following manner:

31
This clause stresses that the precluding of systematic errors (that is, biasedness) by the REH is a
statement rather than a correctly defendable and reasoned finding. Thus there are errors—the only
question is whether they are random ones.
2.6 Unbiasedness: Some Further Considerations 77

covðet ; Ωt1 Þ ¼ 0; ð2:1Þ

where et is the expectation error of period t (as defined above), and Ωt1 denotes the
information set that was exploited when expectations for period t were being
formed in period t  1. However, orthogonality leads to a very important conclu-
sion. Assuming that (for lack of alternatives, this is the only line of reasoning to
adopt)

eti  Ωt1 , where i ¼ 1, 2, 3, . . . ; ð2:2Þ

by doing so we claim that the expectation-forecast errors of t  1 and the preceding

periods (eti) are all included in the information set (Ωt1) available in the period in
which expectations are being formed. In other words, agents had an in-depth
knowledge of errors for the previous periods when expectations were formed for
period t. Based on (2.2.) it also follows from requirement (2.1.) that

covðet ; eti Þ ¼ 0, where i ¼ 1, 2, 3, . . . ; ð2:3Þ

as all previous errors in expectations were already included in the information set
available—in other words, Eq. (2.3) rules out residual autocorrelation. The learning
process is, therefore, completely precluded from the model, since agents will be
incapable of correcting their errors on the basis of their previous errors and, based
on the current model, they will be unable to forecast errors for the next period
(i.e. the next errors of the current model). After all, the absence of opportunity for
learning is caused by the circumstance that, being in possession of the relevant
model, only random (and, in this way, unpredictable) errors may occur—that is,
learning from these errors is both unnecessary and impossible.32

2.6 Unbiasedness: Some Further Considerations

All the details of the previous sections are based on the assumption that knowledge
in the context of the economy which can produce unbiased estimations of factual
outcomes for the future, exists a priori. Simply put, a relevant model is available. In

32
Interestingly, Thomas Sargent and Neil Wallace (1975) deny the possibility of a systematic
economic policy on similar grounds. This is slightly more than refuting its effectiveness, since
even the effort itself seems to be impossible in their reasoning. Aggregate supply can be boosted
only by unexpected increases in the price level generated by the monetary authority. For a surprise
in period t, the non-systematic part of the money supply needs to be unpredictable (the systematic
part is, naturally, always predictable) on the basis of the information set that is available in period
t  1 and that supports forming of expectations for period t. While market agents and the central
bank have access to the same information set, the central bank is unable to set systematically the
random component (that is responsible for the surprise), since the non-correlativeness of the
information available and the unpredictable, random part of the money supply is also true for it.
78 2 The Rational Expectations Hypothesis as a Key Element of New Classical. . .

this case agents can acquire this knowledge by having the relevant information
conveyed to them.33 No other correct, implicit assumption is available, given the
fact that individual expectations of poorly informed agents will probably be
biased.34 This is the first time that we can see something reminiscent of inflation
targeting, as in this regime, agents can utilize the predictions of the quasi-relevant
model (those of the best approximation of the relevant model known to us) with
considerable passivity, of course, since rather than using the model effectively to
compute predictions, agents possess its outputs—and their expectations will thus
coincide with the predictions of the model. It, however, would be a misinterpreta-
tion of the situation to assume that this simplifies the case as bearing in mind the
inherent limitations of the trial-and-error specification when identifying theoretical
parameters, we cannot go any further than confirming the existence of a complete
knowledge which can neither be accessed nor acquired. Therefore, knowledge of
the relevant model is a comfortable and necessary presumption, not a realistic
consequence.
To claim that the relevant model does not exist a priori (it only may exist) is a
starting point that puts the case into a different perspective, i.e. market agents (and
the monetary authority) have to identify it themselves. We cannot circumvent this
problem, since even if the relevant model does exist, it needs to be identified. It was
John Weeks whose critical remarks on the REH were put forth on these grounds
with sharp witticism. This remarkable line of reasoning suggests that the relevant
model cannot be identified by the trial-and-error technique, not even if, theoreti-
cally speaking, a complex and perfect knowledge of the macroeconomic system is
possible. Stochastic data generating process of the economic variable y can be
formulated in the following manner:

yt ¼ x0t β þ ut ; ð2:4Þ

where xt is the vector of all the explanatory variables (excluding random effects, of
course) governing the dynamics of y (containing all relevant information), yt
denotes the numeric value of y in period t, while β is the theoretical parameter
vector and ut is the random term. If there exists a perfect knowledge on a given
economic system, it is obvious that the relevant model implies familiarity with the
theoretical parameters, and our prediction by the model can, therefore, be described
in the following manner:

33
In fact, not only the central bank but any other institution in a central position may be capable of
this. Remember the companies analysing market trends or the ministry of finance. Inflation
targeting did not create expectations, it only recognized their importance and makes attempts to
affect them.
34
It should be borne in mind that market agents here are assumed to be professional
econometricians.
2.6 Unbiasedness: Some Further Considerations 79

Et1 ð yt Þ ¼ x0t β; ð2:5Þ

where Et1 ð:Þ is the prediction of a given variable. Based on this, the prediction
error of the relevant model can be defined by subtracting one equation from the
other, as follows:

yt  Et1 ð yt Þ ¼ ut : ð2:6Þ

The predictions are deviated from actual outcomes by random errors, causing even
the relevant model to be incapable of true predictions. Hence, the formula for
unbiased individual predictions is this:
i
Et1 ð yt Þ ¼ x0t bt1 ; ð2:7Þ

where bt1 is the empirical parameter vector applied for the period t  1 prediction.
Unbiasedness here is a requirement which translates into Eðbt1 Þ ¼ β. If that
requirement is met, i.e. agents are capable of unbiased predictions of actual out-
comes, the expectation error is

yt  Et1
i
ð yt Þ ¼ u^ ti ; ð2:8Þ


since E x0t ½β  bt1  ¼ 0: It has to be noted that errors in the relevant model and
those of individual predictions are not identical. Although random errors may shift
individual estimations from the theoretical parameter vector, these differences will
balance each other out aggregately, causing the theoretical parameter to be the
expected value of the distribution of parameter estimation. However, an estimation
error is derived not only from a specification error, it may equally stem from
contingencies, but its source cannot be identified in specific cases,35 which is
precisely what prevents us from being led to the relevant model empirically. The
model in its intermediate (i.e. incomplete and imperfect) state might still offer an
accurate prediction due to random effects. In that situation, the modeller may,
naturally, assume that his job is done, having acquired some sort of ultimate
knowledge. A considerably more interesting situation is one when the output of
the model is in error in respect of subsequent factual data (it needs to be borne in
mind that these are predictions computed in a stochastic environment). Broadly

35
It should be stressed here that unbiasedness and the circumstance that the expected value of
prediction errors is zero are not interchangeable concepts. As it has been mentioned above,
OLS-estimations always guarantee the mean of errors to be zero. It is still possible in case of a
biased parameter vector that expected value of the prediction error is zero, if the difference vector
of the theoretical and the empirical parameter vector is orthogonal to the vector of the explanatory
variables. However, it is a different problem that, in this case, the prediction error is not
independent of the explanatory variables—a circumstance that can be precluded only by the
unbiasedness of the empirical parameter vector.
80 2 The Rational Expectations Hypothesis as a Key Element of New Classical. . .

speaking, this can be traced back to two possible reasons. In case of a model in an
intermediate state (which Weeks has underlined is the best to be had):
1. The modeller can never identify which part (if any) of the prediction error should
be attributed to the wrong specification (in other words, which is the systematic
part) and
2. Which part reflects random effects.
Hence, he will be unable to improve his model. Moreover, if the perfect model
exists (although based on what has been mentioned above, no account of its
successful specification can be given), errors may occur due to the stochastic
environment (since the Muthian concept of the REH does not require the pre-
dictions to be true), which, once the modeller realized this, he would obviously start
re-specifying his model—which would be a setback compared with the previous
situation. Weeks (1989), while side-lining the apparatus of econometric theory
aside, makes the relevant point that a gradual improvement of the model would
only be possible if we were already familiar with the model which we intend to
approach in a step-by-step manner. The predestined impossibility of the learning
process, coupled with the lack of familiarity by agents of the relevant model, makes
it hard to realistically assume the unbiasedness of estimations. Familiarity with the
relevant model is, literally, a higher level of knowledge, deus ex machina, as it
were, and the impression is that the Walrasian omniscient auctioneer is revived in
the notion of the relevant model.36 Naturally, the fact that the relevant model is
required to “know” the theoretical parameters, might be exposed to criticism.
Seemingly, it necessarily follows from Muth’s phrasing, according to which aver-
ages of expectations are as accurate as complex techniques37 (although at this point
in the reasoning, as it has been noted earlier on, the fact that according to Muth

36
Several parallels can be identified between new classical macroeconomics and the preceding
classical theory from which Filippo Cesarano highlights some, while concentrating on the effects
of anticipated and unanticipated increases of the money supply (Cesarano 1983). Characteristi-
cally, classical economists have also paid more attention to unanticipated increases while virtually
regarding anticipated changes as ineffective (although it is also true that, in the absence of statistics
and information channels almost every change could be labelled as unanticipated—excluding
pre-announced actions). The classicals assumed that, as a fundamental tendency, an increase in the
money supply triggers only gradual adjustments in the price level, but up to that moment real
output and employment also respond (cf. Hume 1752). Imperfect information (and not sticky
prices and wages) played a key role in this story, bearing strong resemblance to the new classical
view. However, what incomplete information is related to it is not an indifferent circumstance. In
this context, the classicals concentrated on changes in the money stock, while new classicals
focussed on the problem of distinguishing relative and absolute changes in the price level.
Cesarano goes as far as referring to the finding that classical economists have implicitly assumed
agents to possess the relevant model, since in his view they could estimate the effects of the
pre-announced actions only by being aware of them.
37
„Averages of expectations in an industry are [. . .] as accurate as elaborate equation systems
[. . .]” (italics added). The fact that “accurateness” is referred to can cause a problem, but it can be
dealt with on the basis of our previous findings (here, it should be regarded as the synonym of
unbiasedness).
2.6 Unbiasedness: Some Further Considerations 81

agents do not engage in econometric modelling is not taken into account). Since, if
agents are supposed to be professional modellers, theoretical parameters will be in
the centre of the probability distribution of unbiased individual estimations, the
numerical values of which are not known to anybody.
If we do not believe in the existence of a complete and perfect knowledge of the
economy, we can only say that, in the absence of this perfect knowledge, every
econometric model can function only as a more or less appropriate approximation
of reality (i.e. of the actual data generating process) and it is evident that market
agents will not be in the same position in any struggle for this knowledge—that is,
individual agents will start their investigation with the likelihood of success being
different for them. Agents with a stronger likelihood for success are offered the
opportunity to approximate the hypothetical relevant model (that does not exist and
cannot be specified at all) more efficiently than others. Here, opportunity means
access to larger and higher quality data sets. At this point, it should be noted that in
the course of this analysis individual agents were assumed to have econometrical
skills characterised by professional standards—in the absence of which the lack of
methodological and technical experience would further reduce the likelihood of
specifying an appropriate model. Mention should also be made of the fact that we
require familiarity with the relevant model itself, and with data sets functioning as
inputs for model estimations by presuming unbiased estimations at the level of both
the relevant model and of individual predictions. Should any of these prerequisites
fail to be met, predictions will appear to be systematically biased.38 The problem of
knowing and identifying theoretical parameters has been discussed earlier on,
although a few additional remarks still need to be made and the issue of having
access to information will be examined at some point later.
Based on mathematical computations, Benjamin Friedman (1979) argues that
not even an infinite timeframe can guarantee that agents will eventually find the
correct model, therefore their expectations will not be rational in the Muthian sense,
i.e. in terms of the hard definition. And now let us re-examine the definition of the
individual expectation error under stochastic conditions:
0
yt  Et1 ð yt Þ ¼ xt ½β  Eðbt1 Þ þ b
ut : ð2:9Þ

The prediction error on the left is independent of the explanatory variables only if
Eðbt1 Þ ¼ β is true, and in this way errors reflect only random effects. In other
words, expectation errors are independent of the explanatory variables only if
parameter estimations are unbiased. So, if β  Eðbt1 Þ 6¼ 0, the expected value of
prediction errors is not zero, which is possible only if yt  Et1 ð yt Þ ¼ b
ut (since the

38
Here, it is worth stressing, yet again, the distinction between the (hypothetical) relevant theory
and individual predictions. We have seen that, in a stochastic environment, identifying the
theoretical parameters is not possible, not even in principle. Therefore, according to the above
interpretation of the REH, there exists a relevant model a priori, which is capable of unbiased
predictions, and the predictions of individual agents (i.e. expectations) are all unbiased estimations
of the data generating process.
82 2 The Rational Expectations Hypothesis as a Key Element of New Classical. . .

error term is a white noise) which, however, is equivalent to β ¼ Eðbt1 Þ, hence

0
xt ½β  Eðbt1 Þ ¼ 0. A biased estimation of model parameters may follow from
various effects: one such factor is the use of a wrong functional form. If the
modeller opts, say, for a linear form, any computed error will be dependent on
previous errors, i.e. serially correlated—and, in this case, it is not true that there is
no prior information for errors, and, by exploiting the knowledge available at the
time of expectations being formed, even the residuals (i.e. prediction errors) will
become predictable albeit with some limitations. Misspecification of model param-
eters occurs also when the estimated parameters of certain significant explanatory
variables are incorrectly presumed a priori to be zero. This case, in its conse-
quences, is equivalent to the forced omission of relevant explanatory variables,
possibly caused by incomplete information. Under these circumstances, prediction
errors reflect the effects of the omitted variables, i.e. errors are predictable on the
basis of these omitted variables and, because of the resulting autocorrelation, the
future dynamics of prediction errors can be approximated by exploiting the infor-
mation in previous residuals (i.e. expectation errors). Meanwhile, the parameters of
the included variables will probably be biased—yet again, the requirement related
to orthogonality fails to be met. The possibility of specifying the relevant model
through the trial-and-error technique might be seen as being further reduced by
these problems—in fact it was known to be impossible right at the outset. Undoubt-
edly, the unbiasedness of individual estimations, if not assumed on an “as if” basis,
implies stringent technical, methodological and information-related requirements.
As has already been pointed out, the availability of information is an equally
important issue. The systematic biasedness of predictions resulting from imperfect
information may still be the case, even if the relevant model exists and its param-
eters are known a priori. As noted previously, agents fall short of receiving equal
opportunities in this race for gathering information: certain data is easily accessible
for everyone, while special information sets are available only for narrow circles of
agents—both data sets may have an impact on the dynamics of a particular
economic variable. Moreover, cost-benefit considerations39 will add another ques-
tion mark to the need to have a full range of information. Inflation forecasts by a
central bank, for example, contain seeds of information that cannot be accessed by
possessing mere data sets. It is a characteristic feature of the Hungarian central bank
routine that central bank modellers modify model outputs on the basis of informa-
tion collected in the course of expert consultations. Several factors that price
dynamics are influenced by cannot be identified by analysing previous time series
data: if, say, a newbie is expected to join the competition of food distributors within
a few months, which present participants are currently aware of (and have estimates
of price policy and competition strategy based on foreign examples, upon which
expectations of the new player will be based), these developments will also have an
impact on current decision as well as those to be taken in the near future. If, for

39
By the way, it is a serious argument for the effectiveness of inflation targeting, since it may be
more economical for actors to adapt and use ready-made model outputs.
2.7 Consequences: The Road to Inflation Targeting 83

example, the new competitor is expected to use lower prices in order to reshape
market share conditions, even current price dynamics will change, i.e. competitors
will bend over backwards to keep prices at bay in order to prepare for keener
competition. What the central bank does by modifying its model predictions on the
basis of this information is nothing more than form rational expectations. However,
it should also be noted that, generally speaking, central bank modelling gives
different priorities to different estimation methods with regards to the relevant
time horizon. In the short run, expert predictions are very useful, i.e. model outputs
are modified on this basis, while model predictions become increasingly significant
on a lengthened time horizon. Equally important is the fact that the predictions of
agents in need of additional information will deviate from the central bank forecast,
despite the fact that both are rational (at least in weak definition terms) as infor-
mation unknown (and almost unavailable) cannot be taken into account by the
actors.40 Still, it is unrealistic to assume that market agents forced to do without
expert consultations have equal access to knowledge of this kind, for the simple
reason that if this knowledge was readily available, central bank economists
(as everyday consumers) themselves would have access to it and expert-level
consultations would become useless. It is, in fact, unnecessary as Muth’s definition
itself stopped short of prescribing, that the relevant model (substituted here by a
central bank forecast) and individual predictions should be identical. However, it
follows from this reasoning that not even the central bank model is void of bias,
let alone individual predictions.

2.7 Consequences: The Road to Inflation Targeting

Our reasoning so far has been focusing on highlighting the realization that the
alternatives of specifying the relevant model are limited, and the full and perfect
knowledge of a macroeconomic system can be regarded as a comfortable presump-
tion at best (albeit not a realistic one). Knowledge of the theoretical parameter
vector is not feasible, even if high-level econometrical techniques are accessible or
an adequate identification of the relevant explanatory variables is performed—all
the more so given that the unavailability of certain types of information makes the
task of modelling even more difficult to accomplish. Previously, a new concept (or,
to be more precise, a characteristic) was inconspicuously introduced. It says that the
biasedness of estimations is not independent of the quantity and quality of infor-
mation to which the modeller has access. In other words, estimations are asymp-
totically unbiased with respect to the exploited information, i.e.:

40
Barro and Fischer (1976) use the concept of rationality of expectations by providing an
interpretation that is very similar as they regard as rational every forecast that can be considered
as an optimal estimation of future outcomes on the basis of available information—even if a
market agent does not have access to all the information that is exploited in the model.
84 2 The Rational Expectations Hypothesis as a Key Element of New Classical. . .

 
b ¼ Θ;
lim E Θ ð2:10Þ
i!i*

where Θ is the theoretical parameter, Θb is its empirical estimation, and i denotes the
amount of information processed while estimations were being made. If the latter
approaches full and perfect knowledge (i*), empirical (that is, estimated) parame-
ters will continue to move closer to the theoretical parameters of the data generating
process. Consistency can be interpreted similarly:
 
b  Θ þ ε ¼ 1, for all ε > 0:
lim P Θ  ε  Θ ð2:11Þ
i!i*

On the basis of asymptotical unbiasedness, a moderated41 version of the strong

definition of the REH can be suggested in which familiarity with the theoretical
parameters is not a requirement for the relevant model—which, of course, implies
that the assumption on the unbiasedness of individual estimations is no longer
considered valid. An agent (e.g. the central bank) with access to information along
with methodological skills, which are satisfactory in quantitative and qualitative
terms alike, will be able to specify a quasi-relevant model (by increasing the
quantity of information processed it will be capable of reducing biasedness)
describing a specific macroeconomic system, and the more focal its position, the
stronger the likelihood to make other agents accept it. All we do is ease the
assumptions and what (i.e. mere unbiasedness and not the knowledge of the
theoretical parameter vector) has thus far been required from individual estimations
is now a requirement imposed on the quasi-relevant model. If, as a result, individual
estimations turn out to be biased, averages of individual estimations will no longer
give the relevant model. We need to take only one step to move towards the
opportunity of successfully introducing inflation targeting regimes as agents42
will decide to adjust their expectations to the outputs of the (quasi) relevant
model. In this way, individual expectations become unbiased as much as possible,
albeit against a background of considerable passivity. If this was not the case, the
central bank would be in a position to control the output through deception—and
might only refrain from doing this because it is deemed unnecessary, given the
manner in which its objection function is specified.43
We have seen that the unbiasedness of individual estimations is likely to follow
directly from the circumstances that characterize modelling, therefore if we want to
interpret rational expectations on the basis of asymptotical unbiasedness with
respect to information, an appropriate specification of the objection function is

41
This moderate version needs to be clearly distinguished from the weak definition, since we
remain within the scope of the hard definition when using this modified form.
42
Upon realizing that their efforts to process information have proved useless, since the outcome
from those efforts has been limited to systematically biased estimation.
43
As it will be seen in Chap. 6, in inflation targeting regimes this can be avoided by the central
bank’s commitment to keeping inflation at a low level.
2.7 Consequences: The Road to Inflation Targeting 85

needed in order for deception-based central bank policies to be precluded. The

explanation of inflation targeting can be derived from the modified version of the
REH (i.e. one based on the adoption of outputs from the quasi-relevant model),
given the central bank’s position to control expectations and its commitment to
avoid time-inconsistent policies, which prevents it from using potential trade-offs.
However, it should be understood that, as for the definition of the REH, inclusive of
the interpretation as Muth meant it to be, there would be no room for manoeuvre in
this concept for a central bank to exert influence on expectations, since actors
forming rational expectations can make unbiased estimations of their own.
It seems, therefore, that the information side of the bias in aggregated individual
estimations can equally be understood upon reflection. It goes without saying that
estimations are specific functions stemming from information sets adopted. It is not
too much of a challenge to recognize the existence of widely and easily accessible
information, while some information can be labelled as insider data with a narrower
circle of agents. Let us take an example. The assumption is that a national economy
is comprised of two distinct groups of agents: by adopting the above example, one
group is that of food product consumers, the other group is a distributor of food
products (from the perspective of that example, the sector itself is completely
indifferent). Both the consumers and the dealer are individual agents. In this case,
consumers form their inflationary expectations solely on the basis of past experi-
ences, whereas the dealer takes into consideration all its special knowledge as well,
which, in the context of inflation dynamics, is also relevant information. Unbiased-
ness means that future data is couched in individual predictions in such a manner
that biases of individual predictions level off—however, a situation like this will
only be brought about by accident as a strongly over-represented information set,
used in the forming of expectations, can be identified at an aggregate level.44 By
assuming that an agent, having accessed special information (available only to a
limited group of agents), is also aware of the knowledge (on past dynamics)
acquired by consumers, his prediction will move closer to (later) outcomes
(or the expected value of the probability distribution of outcomes), but in order to
achieve aggregate unbiasedness, all other predictions (biased equally and in the
same direction as has been described in the example) would need to be biased in a
way that these shortcomings will balance each other out. However, this is not
possible because biases point consistently in the same direction—moreover, instead
of a compensation, aggregated expectations will either underestimate or
overestimate factual data. The shortcomings of a prediction, also based on special
information,45 would need to be levelled out by all other estimations in order to
form unbiased expectations on the average—while there is no reasoned principle to
support the acceptance of this clause as a presumption (unless we are ready to do
this in an arbitrary manner). Unbiasedness at the aggregate level (i.e. a condition in
which a later outcome or the theoretical parameter vector appears as the average of

44
Being over-represented means that it is taken into account as a significantly high priority.
45
This is still not the relevant or quasi-relevant model.
86 2 The Rational Expectations Hypothesis as a Key Element of New Classical. . .

individual estimations) may emerge accidentally at best, and even if such a situation
occurs (i.e. estimations are unbiased at the aggregate level), its recurrent occurrence
should not be expected in any way. Unbiased aggregated estimations are unlikely to
result from biased individual estimations—something not even Muth denied given
that his definition prioritized the amount of information available. However, this
informational symmetry can only be assumed with some difficulty as biased
individual estimations might be the outcome of various factors. An additional
problem is that certain assumptions in the REH are not consistent with each
other, since unbiasedness as opposed to expectation formation assumed to be
based on non-professional techniques, preclude each other (here the as-if-reasoning
renders useless the theory: how can cross-sectional differences in opinion be
explained when expectations are formed in a manner similar to that of the relevant
theory) and the estimation of the relevant model, based on familiarity with the
theoretical parameters, is an almost insurmountable task if undertaken empirically.
All this eventually takes us to inflation targeting from Muth’s hard definition of
the REH. To get to this point, all we needed to do was to realize that agents cannot
be assumed to make unbiased estimations of future outcomes, unless an as-if-
reasoning is adopted. However, assumptions of the REH can be investigated not
only as a pure theory but on grounds of the intention to describe reality in a realistic
way. As we have seen, for reasons due to methodological difficulties, knowledge on
the theoretical parameter vector cannot be acquired via the traditional trial-and-
error technique. Even if it is accepted that agents can have outstanding macro-
modelling skills, individual estimations are still biased due to limited accessibility
to information, thereby creating an opportunity for a systematic monetary policy
based on deception. According to the moderate version of the REH introduced here,
a sufficient degree of unbiasedness of expectations is facilitated by circumstances in
which market agents adopt the outputs of the model that performs the best (i.e. of
the quasi-relevant model), and in this way predictions of the model and those of the
agents will coincide. In the absence of this, agents can only make biased estima-
tions, making a systematic countercyclical central bank policy possible. If, follow-
ing the moderate version, the hard definition of the REH is interpreted in terms of
inflation targeting, the need for prohibition on deception will need to be taken into
account: a central bank capable of and ready for commitment can anchor expecta-
tions, since it does not support short-run output maximization by controlling
expectations. Since the unbiasedness of individual estimations (or, to use a cau-
tiously worded language: the absence of biasedness reaching the greatest possible
extent) is not the consequence of the performance of individual modellers, the lack
of deception by the central bank can be interpreted only as a voluntary commitment.
Without this the functionality of inflation targeting would be jeopardized.
There is yet another circumstance to highlight. For new classicals, the relevant
model is the framework investigated in Chaps. 4 and 5.46 The question is, and this
book itself has been devoted to the way in which this problem should be addressed,

46
This important aspect was brought to my attention by Prof. Lászl
o Vı́gh in a conversation.
2.7 Consequences: The Road to Inflation Targeting 87

whether an economist who is not a follower of the new classical school of thought,
should accept the relevancy of this model. The line of reasoning that we follow is
that this is feasible only with serious caveats. It seems, at the same time, that the
mechanism of inflation targeting is described well enough by the hard definition of
the REH if (as is clear from the above details) the rigour of assumptions is loosened
and we rest satisfied with a setting in which the relevant model is transmitted to the
agents by the central bank. Moreover, one does not need to be a monetarist to
believe in the functionality of this mechanism: a relevant model is not necessarily
motivated by a monetarist approach. Sure enough, another problem needs to be
dealt with: in an orthodox Keynesian model, for example, price dynamics is
explicitly defined, leaving no room for expectations. Therefore, not every macro-
model will be consistent with the mechanism based on communication by the
central bank, not every model can be the quasi-relevant model in the focus of
central bank communication—except those (kinds of) models that derive price
changes from expectations (or those that leave room for expectations in price
dynamics). What a central bank thinks about its own transmission mechanism
does carry significance. These problems will only present themselves if the new
classical theory is not thought of as a monolithic mass but rather as a puzzle. This is
the only case where the relevant model may be something else than the new
classical construct.
There is no easy answer to some questions. Is the realization that even the quasi-
relevant model of the central bank is not necessarily unbiased conducive to the
finding that public expectations anchored by that model are also biased? Should we
infer from this that the central bank anchors expectations erroneously, which would
lead to a systematic bias? If the rationality of expectations is disregarded, the
answer is that the introduction of the quasi-relevant model of the central bank
may cause public expectations to have systematic biasedness. This shortcoming
cannot be fully eliminated by using the trial-and-error technique, therefore, the
quasi-relevant model will cause a (prima facie) disequilibrium state. Whether real
output will fall below or rise above the potential (equilibrium) level for longer
periods or if these cycles will be only random pendulum swings—it will hinge on
the nature of expectation errors. However, this very quality of errors is what should
be compared to the (alleged) rationality of expectations. Should any kind of
systematic error occur, it will be recorded by the central bank modellers and the
public alike and, accordingly, model predictions will be modified—this was exactly
what we pointed out earlier on in the context of expert predictions. Errors of the
quasi-relevant model can, therefore, only be labelled random, resulting, among
other things, in the volatility of real output, which is higher than what would be
experienced on the basis of the relevant model.
It seems that new classical macroeconomics is capable of giving us considerable
assistance in understanding the mechanisms of inflation targeting. Economists,
reading the pertinent literature, may be inclined to think that there is considerable
experience and technical knowledge behind inflation targeting regimes. This is true,
although that knowledge is mainly focused on direct, factual and observed mech-
anisms (i.e. interactions). Hardly any mention is made of principles that underscore
88 2 The Rational Expectations Hypothesis as a Key Element of New Classical. . .

and, as it were, facilitate the functionality of inflation targeting. Literature on

inflation targeting has a distinct concept of humans, while reflecting clear assump-
tions on the structure of macro-systems and the motivations of agents—while
turning a blind eye to the fact that these elements are not axioms. The particulars
of this system of presumptions can be modulated, its existence might be argued for
or against and, seemingly, it can be traced back to underlying premises. It is not
enough to assert that central banks are capable of anchoring public expectations—it
also needs to be explained why they are capable of this. That is where new classical
macroeconomics and the REH prove helpful. It is precisely what has been proved in
Chap. 2: market agents are incapable of doing the job of central banks. Based on
this alone, though, rationality cannot be argued against, since the availability of the
most reliable prediction is of crucial importance from the point of view of economic
activity. Rather than questioning the rationality of agents due to their shortage of
skills, an institution needs to be introduced—one capable of adopting functions
such as modelling and forecasting. If market agents are committed to forming
rational expectations, we have serious arguments to support the idea that expecta-
tion formation is suboptimal and the adoption of easily accessible data capable of a
satisfactory performance will be preferred to individual procedures. In this way, the
result can be identical with what is required by new classicals: the probability
distribution of market expectations will coincide with the outputs of the (quasi)
relevant model. Eventually, inflation targeting results in a partial manifestation of
the essential new classical doctrines.

References

Anderson RG, Kavajecz KA (1993) A historical perspective on the Federal Reserve’s monetary
aggregates: definition, construction and targeting. Federal Reserve Bank of St Louis Review 77
(3–4):1–31
Arrow KJ (1973) Information and economic behaviour. Federation of Swedish Industries,
Stockholm
Babbie E (1989) The practice of social research. Wadsworth, Belmont, Hungarian edition: Babbie
E (2000) A társadalomtudományi kutatás gyakorlata (trans: Kende G, Szaitz M). Balassi,
Budapest
Barro RJ, Fischer S (1976) Recent developments in monetary theory. J Monet Econ 2(2):133–167
Begg DKH (1982) The rational expectations revolution in macroeconomics–theories and evi-
dence. Philip Allan, Oxford
Berlin I (1958) Two concepts of liberty. In: Berlin I (ed) (1969) Four essays on liberty. Oxford
University Press, Oxford, pp 118–172
Bronk R (2009) The romantic economist–imagination in economics. Cambridge University Press,
Cambridge
Bryant J (1983) A simple rational expectations Keynes-type model. Q J Econ 98(3):525–528
Bungard W (1997) A proszociális magatartás empirikus kutatásának m odszertani problémái. In:
Lengyel Z (ed) Szociálpszichologia. Osiris, Budapest, pp 23–36
Caginalp G, McCabe K, Porter D (2003) The foundations of experimental economics and
applications to behavioral finance–the contributions of Nobel laureate Vernon Smith. J
Behav Finance 4(1):3–6
References 89

Cesarano F (1983) The rational expectations hypothesis in retrospect. Am Econ Rev 73

(1):198–203
Dickinson DG, Driscoll MJ, Ford JL (1982) Rational expectations, random parameters and the
non-neutrality of money. Economica 49(195):241–248
Elster J (1989) Nuts and bolts for the social sciences. Cambridge University Press, Cambridge,
Hungarian edition: Elster J (2001): A társadalom fogaskerekei (trans: Helmich Z, Szánt o Z).
Osiris, Budapest
Erdős T (1998) Inflácio. Akadémiai, Budapest
Farkas Z (2006) A racionális cselekvés, a társadalmi cselekvés és k€ olcs€
onhatás. Miskolci
Egyetem, Miskolc
Fischer S (1977) Long-term contracts, rational expectations, and the optimal money supply rule. J
Polit Econ 85(1):191–205
Friedman B (1979) Optimal expectations and the extreme information assumptions of ’rational
expectations’ macromodels. J Monet Econ 5(1):23–41
Friedman M (1953) The methodology of positive economics. In: Friedman M (ed) Essays in
positive economics. University of Chicago Press, Chicago, pp 3–43
Friedman M (1968) The role of monetary policy. Am Econ Rev 58(1):1–17
Friedman M (1970) The optimum quantity of money and other essays. Aldine Publishing Com-
pany, Chicago
Gerrard B (1994) Beyond rational expectations: a constructive interpretation of Keynes’s analysis
of behaviour under uncertainty. Econ J 104(3):327–337
Hahn FH (1980) Monetarism and economic theory. Economica 47(185):1–17
Harasztosi P (2004) Racionális várakozások és a Monetáris Konszenzus. Fordulat (Spring
2004):64–75
Hoover KD (1990) The new classical macroeconomics: a sceptical inquiry. Blackwell, Oxford
Horváth L (2000) Monetarizmus. In: Bekker Z (ed) Alapművek, alapirányzatok, vol II. Aula,
Budapest, pp 433–455
Hume D (1752) Of money. In: Hume D (ed) (1994) Political essays. Cambridge University Press,
Cambridge, pp 115–125
Jakab MZ, Vadas G (2001) A háztartások fogyasztásának előrejelzése € okonometriai
m odszerekkel. Central Bank of Hungary, Budapest
Kantor B (1979) Rational expectations and economic thought. J Econ Lit 17(4):1422–1441
Keuzenkamp HA (1991) A precursor to Muth: Tinbergen’s 1932 model of rational expectations.
Econ J 101(9):1245–1253
Király J (2000) Robinson és a k€ozgazdászok. Café Bábel 10(2):107–113
KVA (2002) Foundations of behavioral and experimental economics–Daniel Kahneman and
Vernon Smith (Laudation). The Royal Swedish Academy of Sciences, Stockholm
Maddala GS (2001) Introduction to econometrics, 3rd edn. Wiley, New York, Hungarian edition:
Maddala GS (2004) Bevezetés az €okonometriába (trans: Békés G, Cseres-Gergely Z, Csorba G,
Reiff Á, Szentpéteri Á, Szilágyi K). Nemzeti Tank€onyvkiado, Budapest
Malhorta NK, Simon J (2009) Marketingkutatás. Akadémiai, Budapest
Mátyás A (1984) A polgári k€ozgazdaságtan napjainkban. Kossuth, Budapest
Mill JS (1859) On liberty, 2nd edn. John W. Parker & Son, London
Mises L (1981) The theory of money and credit. Liberty Classics, Indianapolis
Muth JF (1961) Rational expectations and the theory of price movements. Econometrica 29
(3):315–335
Pete P (2001) Bevezetés a monetáris makro€okonomiába. Osiris, Budapest
Porter DP, Smith VL (2003) Stock market bubbles in the laboratory. J Behav Finance 4(1):7–20
Ramanathan R (2002) Introductory econometrics with application, 5th edn. South-Western, New
Delhi, Hungarian edition: Ramanathan R (2003) Bevezetés az € okonometriába alkalmazásokkal
(trans: Marcinkovics R, Nagy G, Petrimán Z). PANEM, Budapest
Rozgonyi T, Toledo R (2008) Halmazok és függvények. Nyı́regyházi Főiskola, Nyı́regyháza
90 2 The Rational Expectations Hypothesis as a Key Element of New Classical. . .

Sargent TJ, Wallace N (1975) “Rational” expectations, the optimal monetary instrument, and the
optimal money supply rule. J Polit Econ 83(2):241–254
Shaw GK (1984) Rational expectations–an elementary exposition. Harvester Press, Brighton
Smith VL (1976) Experimental economics–induced value theory. Am Econ Rev 66(2):274–279
Smith VL (1982) Markets as economizers of information–experimental examination of the
“Hayek Hypothesis”. Econ Inq 20(2):165–179
Smith VL (1989) Theory, experiment and economics. J Econ Perspect 3(1):151–169
Smith VL (1991) Rational choice–the contrast between economics and psychology. J Polit Econ
99(4):877–897
Smith VL (1994) Economics in the laboratory. J Econ Perspect 8(1):113–131
Smith VL (2003) Nobel lecture: Constructivist and ecological rationality in economics. Am Econ
Rev 93(3):465–508
Smith VL, Suchanek GL, Williams AW (1988) Bubbles, crashes, and endogenous expectations in
experimental spot asset markets. Econometrica 56(5):1119–1151
Thornton S (2009) Karl Popper. In: Zalta EN (ed) The Stanford encyclopedia of philosophy,
Summer 2009 edition. http://plato.stanford.edu/entries/popper. Accessed 11 Nov 2010
T
oth IJ (2002) Vállalati és lakossági konjunktúra felmérések Magyarországon. Central Bank of
Hungary, Budapest
T
oth IJ, Vincze J (1999) Hihetünk-e a vállalati felvételeken alapulo kutatások eredményeinek?
Stat Szle 77(10–11):844–855
Vadas G (2001) Túl a makro-változokon: a lakossági bizalmi index és a magyar háztartások
fogyasztási kiadása. Central Bank of Hungary, Budapest
Varian HR (1999) Intermediate microeconomics. W.W. Norton, New York/London, Hungarian
edition: Varian HR (2001) Mikro€okonomia k€ozépfokon–egy modern megk€ ozelı́tés (trans:
Bara Z, Temesi J). KJK–KERSZÖV, Budapest
Várpalotai V (2003) Disaggregated cost pass-through based econometric inflation-forecasting
model for Hungary. Central Bank of Hungary, Budapest
Vigvári A (2004) Pénzügy(rendszer)tan. KJK–KERSZÖV, Budapest
Visco I (1984) Price expectations in rising inflation. North-Holland, Amsterdam/New York/
Oxford
Weeks J (1989) A critique of neoclassical macroeconomics. Macmillan, London, Hungarian
edition: Weeks J (1998) A neoklasszikus k€ozgazdaságtan kritikája (trans: Szalai Z). Aula,
Budapest
http://www.springer.com/978-3-319-17577-5