This action might not be possible to undo. Are you sure you want to continue?

Welcome to Scribd! Start your free trial and access books, documents and more.Find out more

www.econstor.eu

Der Open-Access-Publikationsserver der ZBW – Leibniz-Informationszentrum Wirtschaft

The Open Access Publication Server of the ZBW – Leibniz Information Centre for Economics

Nutzungsbedingungen:

Die ZBW räumt Ihnen als Nutzerin/Nutzer das unentgeltliche,

räumlich unbeschränkte und zeitlich auf die Dauer des Schutzrechts

beschränkte einfache Recht ein, das ausgewählte Werk im Rahmen

der unter

→ http://www.econstor.eu/dspace/Nutzungsbedingungen

nachzulesenden vollständigen Nutzungsbedingungen zu

vervielfältigen, mit denen die Nutzerin/der Nutzer sich durch die

erste Nutzung einverstanden erklärt.

Terms of use:

The ZBW grants you, the user, the non-exclusive right to use

the selected work free of charge, territorially unrestricted and

within the time limit of the term of the property rights according

to the terms specified at

→ http://www.econstor.eu/dspace/Nutzungsbedingungen

By the first use of the selected work the user agrees and

declares to comply with these terms of use.

zbw

Leibniz-Informationszentrum Wirtschaft

Leibniz Information Centre for Economics

Ein Mitglied der

Lengnick, Matthias; Wohltmann, Hans-Werner

Working Paper

Agent-based financial markets and

New Keynesian macroeconomics: A

synthesis

Economics working paper / Christian-Albrechts-Universität Kiel, Department of Economics,

No. 2011,09

Provided in cooperation with:

Christian-Albrechts-Universität Kiel (CAU)

Suggested citation: Lengnick, Matthias; Wohltmann, Hans-Werner (2011) : Agent-based

financial markets and New Keynesian macroeconomics: A synthesis, Economics working

paper / Christian-Albrechts-Universität Kiel, Department of Economics, No. 2011,09, http://

hdl.handle.net/10419/49990

agent-based financial markets and new

keynesian macroeconomics -a synthesis-

-updated version-

by Matthias Lengnick and Hans-Werner Wohltmann

No 2011-09

Agent-Based Financial Markets and

New Keynesian Macroeconomics

– A Synthesis –

Matthias Lengnick

∗

University of Kiel

University of Mannheim

Hans-Werner Wohltmann

∗∗

University of Kiel

September 13, 2011

Abstract

We combine a simple agent-based model of ﬁnancial markets and a New Keynesian macroeconomic

model with bounded rationality via two straightforward channels. The result is a macroeconomic

model that allows for the endogenous development of business cycles and stock price bubbles. We

show that market sentiments exert important inﬂuence on the macroeconomy. They introduce high

volatility into impulse-response functions of macroeconomic variables and thus make the eﬀect of

a given shock hard to predict. We also analyze the impact of diﬀerent ﬁnancial transaction taxes

(FTT, FAT, progressive FAT) and ﬁnd that such taxes can be used to stabilize the economy and

raise funds from the ﬁnancial sector as a contribution to the costs produced by the recent crisis. Our

results suggest that the FTT leads to higher tax revenues and better stabilization results then the

FAT. However, the FTT might also create huge distortion if set too high, a threat which the FAT

does not imply.

JEL classiﬁcation: E0, E62, G01, G18

Keywords: Agent-based modeling; stock market; New Keynesian macroeconomics; ﬁnancial trans-

action tax; ﬁnancial activities tax

∗

E-mail: matthias.lengnick@vwl.uni-mannheim.de, Tel: +49 621 181 1897, Fax: +49 621 181 1774

∗∗

E-mail: wohltmann@economics.uni-kiel.de

Economists [...] have to do their best to incorporate

the realities of ﬁnance into macroeconomics.

Paul Krugman (2009)

1 Introduction

The economies of almost every country have recently been hit by a turmoil in the ﬁnancial markets.

This so-called ﬁnancial crisis has vividly demonstrated that developments in the ﬁnancial markets

can have major impacts on the real economy. Interdependencies between real and ﬁnancial markets

should therefore obviously be taken into account when doing macroeconomics. Natural questions

to ask after the recent crisis are: To which extent does the formation and bursting of bubbles spill

over into real markets? Can ﬁnancial market regulation be used to reduce disturbances of the real

economy? How can the ﬁnancial sector be hold to account for the enormous costs created by the

recent crisis?

For about two decades now, a relatively new modeling approach has been applied to the analy-

sis of ﬁnancial and foreign exchange markets. This approach builds on the method of agent-based

computational (ABC) simulation, it drops the assumptions of rational expectations, homogeneous

individuals, perfect ex ante coordination and often also market equilibria, in favor of adaptive learn-

ing, simple interactions of heterogeneous agents, and emerging complex macroscopic phenomena.

1

The approach seems very promising thus far since, on the one hand, it is grounded in the results of

survey studies

2

and laboratory experiments

3

, and on the other hand, the emerging macro-dynamics

mimic the properties of real world data (such as martingale property of stock prices, fat tails of

return distribution, volatility clustering and dependency in higher moments)

4

quite well, a success

that traditional ﬁnancial market models, building on equilibrium and rationality, do not provide.

5

A huge literature has already developed on this topic that – despite its success – is largely ignored

by macroeconomists.

1

For an introduction into ABC ﬁnancial market modeling see, e.g., Samanidou et al. (2006), Hommes (2006) or LeBaron

(2006). Outstanding examples of such models are Kirman (1993), Brock and Hommes (1998), and Lux and Marchesi

(2000).

2

Consult Frankel and Froot (1987), Ito (1990), Taylor and Allen (1992) and Lui and Mole (1998).

3

Consult Caginalp et al. (2001), Sonnemans et al. (2004) and Hommes et al. (2005).

4

A detailed description of these stylized facts can be found in Lux (2009).

5

De Grauwe and Grimaldi (2006), for example, compare the performance of an agent-based model with popular models

like that of Obstfeld and Rogoﬀ in explaining the stylized facts of foreign exchange rates. They ﬁnd that the former

performs much better.

1

One strength of the ABC method is that it naturally allows for the endogenous emergence of

bubbles. In such models, investors can typically choose from a set of diﬀerent non-rational trading

strategies. The decision which strategy to employ is reached by using an evolutionary approach: A

continuous evaluation of those strategies according to past performance leads to changes in the size

of the diﬀerent investor groups. In phases that are dominated by technically operating investors,

stock prices can deviate sharply from their underlying fundamental value. If market sentiments

change and fundamentalists dominate, convergence towards the fundamental value sets in. Inspired

by the spectacular failure of mainstream macroeconomics to provide an explanation of the current

crisis and an agenda of how to deal with it, a number of authors are calling for the use of ABC

models in macroeconomics.

6

According to them, the assumptions of equilibrium, perfect ex ante

coordination, rational expectations and representative agents are very unrealistic and the reason

that macroeconomists have become blind to crisis.

The emergence of asset price misalignments (i.e. bubbles) on the ﬁnancial markets is often seen as

having the most devastating impact on the real economy. Some macroeconomic models already allow

for such misalignments. Bernanke and Gertler (1999), for example, augment the model of Bernanke

et al. (1999) by imposing an exogenously given path for asset price misalignment. In their model,

each bubble has a constant exogenous probability to burst, where ”burst” simply means that asset

prices immediately return to their fundamental value. More recently Milani (2008) and Castelnuovo

and Nistico (2010) have integrated stock price misalignment into a New Keynesian DSGE model.

Their aim is to provide insights into the dynamics of the stock price component that is driven by

utility-optimizing, rational-expecting agents. Stock price dynamics in such models are a rational

response of an ex ante perfectly coordinated economy in equilibrium. With the mentioned criticism

in mind, it is hard to imagine stock price bubbles or ﬁnancial crisis in such frameworks.

Kontonikas & Ioannidis [KI] (2005) and Kontonikas & Montagnoli [KM] (2006) use forward- and

backward-looking New Keynesian macroeconomic (NKM) models with lagged stock wealth eﬀects.

Stock price dynamics in these models are not exogenously imposed and the crash of a bubble does not

simply occur with a ﬁxed probability. Instead they make use of an endogenous dynamic process that

binds stock prices to two diﬀerent forces: One of which leads to a return towards the fundamental

6

See, e.g., Colander et al. (2008), Colander et al. (2009), Lux and Westerhoﬀ (2009), Krugman (2009), Kirman (2010),

Delli Gatti et al. (2010), and Dawid and Neugart (forthcoming). Examples of purely agent-based macro models (with

no connection to NKM) are Gaﬀeo et al. (2008) or Deissenberg et al. (2008).

2

value, and the other – so-called momentum eﬀect – relates stock prices to their own past development.

While KI (2005) and KM (2006) are clearly inspired by the agent-based ﬁnancial markets literature

with its fundamentalist and chartist trading rules, none of the above models explicitly motivates the

dynamics of stock price misalignment by boundedly rational investor behavior and none makes use

of an evolutionary selection mechanism of trading strategies that is used in ABC type models.

In a recent paper Bask (2009) uses a New Keynesian dynamic stochastic general equilibrium

(DSGE) framework with stock prices that are determined by the demand of two diﬀerent types of

investors: chartists and fundamentalists. While the model provides the major advantage that it

justiﬁes stock price movements by the behavior of these two types of investors, it does not allow for

an endogenous evaluation of the diﬀerent investment strategies. As in KI (2005) and KM (2006) the

aspect of evolutionary learning, that is so important for ABC ﬁnancial markets, is missing. Investors

therefore keep employing the same investment rule and do not try to learn from past observations.

In this paper, we merge a simple ABC model of ﬁnancial markets with the New Keynesian DSGE

model. Such a comprehensive model allows for the simultaneous development of endogenous business

cycles and stock price bubbles. Expectations in both submodels are formed by an evolutionary

selection process. To the best of our knowledge, no such attempt has been made so far. Since we

combine two separate subdisciplines of economics, and do not want to exclude readers who are not

familiar with both of these areas, our approach focuses on simplicity. Nonetheless, our model leads

to a number of interesting insights. We ﬁnd that the transmission of shocks is dependent on the

state of market sentiments

7

at the time of its occurrence. We also ﬁnd that the negative impact that

speculative behavior of ﬁnancial market participants exerts on the macroeconomy, can be reduced

by introducing a tax on ﬁnancial transactions. We use our model to answer two questions that

are currently on the international policy agenda: (1) Is a Financial Transaction Tax (FTT) or a

Financial Activities Tax (FAT) better suited for regulating ﬁnancial markets and generating tax

income for the state? (2) Should the rate of a FAT be ﬂat or progressive?

The model is developed in section 2. We demonstrate the working of our model by means

of numerical simulation and impulse response analysis in section 3. In section 4 we analyze the

7

By market sentiments we mean the state of agents opinions about economic variables that are the result of an evolu-

tionary process and not of rational forecasting. The expression is taken from De Grauwe (2010a) where it is used as a

synonym to animal spirits.

3

introduction of diﬀerent kinds of taxes levied on ﬁnancial transactions. Our results are checked for

robustnes in section 5. Section 6 concludes.

2 The Model

Our model consists of two parts, one describing the ﬁnancial sector, and one the real sector of the

economy. We use the ABC chartist-fundamentalist model proposed by Westerhoﬀ (2008) to model

the ﬁnancial market. The real sector is described by the NKM framework augmented by a cost eﬀect

of stock prices. Since we allow for an endogenous development of business cycles and stock price

bubbles, our model is an augmentation of NKM models that already include stock price bubbles,

but impose their dynamics exogenously (section 1). It is also an augmentation of those models that

integrate a stock market with diﬀerent types of investors into macroeconomics, but do not employ

endogenous learning (section 1).

An approach which is related to ours can be found in Proa˜ no (2011). The author makes use of an

ACE foreign exchange market in the context of an open economy macro model. While the general

idea is similar to ours, our way of integrating the ﬁnancial market and the real economy is very

diﬀerent. Our ﬁnancial market, in contrast to Proa˜ no (2011), ﬁrst, contains noise and is therefore

not completely deterministic. Second, it operates on a smaller time interval than the real economy.

The paper at hand can thus be seen as a complementary approach to a similar research target.

The ﬁrst problem one has to deal with is that the rules determining the dynamics of ﬁnancial

markets are likely to be very diﬀerent from those of the real markets. Economic transactions in the

former seem to take place much more frequently than in the latter.

8

For example, a large fraction of

ﬁnancial transactions (10%-60% according to market)

9

are accounted by algorithmic trading which

is typically of an extremely short-term intra-daily nature. This implies that both can not be modeled

on the same time scale.

10

The two modeling methodologies employed throughout this paper are building on very diﬀerent

assumptions. In order to prohibit contradictions stemming from these diﬀerent assumptions of

8

Although this argument seems to be straightforward it is also backed empirically by Aoki and Yoshikawa (2007), who

ﬁnd that time series of real economic data do not share the power law distribution of ﬁnancial markets which implies

that the latter are characterized by higher economic activity.

9

Consult Matheson (2011), p. 19.

10

On the explicit modeling of high frequency New Keynesian models see Franke and Sacht (2010).

4

ABC and DSGE modeling, we do not simply integrate one into the other, but take the diﬀerences

seriously. As a result, we must assume that real and ﬁnancial markets are populated by diﬀerent

kinds of agents. We interpret those of the ﬁnancial market to be institutional investors, who have

the resources to participate in high frequency trading. Conversely, real market agents have neither

detailed knowledge about ﬁnancial markets, nor the possibility to participate in high frequency

trading. Subsection 2.1 deﬁnes the ﬁnancial sector of our economy, while 2.2 deﬁnes the real one.

Subsection 2.3 brings the two sectors together.

2.1 Financial Market

We use the model proposed by Westerhoﬀ (2008) to deﬁne the ﬁnancial sector of our economy for

two reasons: First, because of its straightforward assumptions and easy implementation, and second,

because it has already been used for policy analysis (especially transaction taxes) so that its behavior

in this respect is well known.

11

In this model, stock price adjustment is given by a price impact

function:

s

t+1

= s

t

+ a

_

W

C

t

D

C

t

+ W

F

t

D

F

t

_

+ ǫ

s

t

(1)

D

C

and D

F

stand for the orders generated by chartists and fundamentalists, respectively.

12

W

C

and W

F

denote the fractions of agents using these strategies, and a is a positive reaction parameter.

Eq. (1) can be interpreted as a market maker scenario, where prices are adjusted according to

observed excess demand.

13

Since fundamentalist and chartist investment strategies do not account

for all possible strategies that exist in real markets, a noise term ǫ

s

is added that is i.i.d. normally

distributed with standard deviation σ

s

. It can be interpreted as the inﬂuence of those other strategies.

t denotes the time index which is interpreted as days. For the sake of simplicity, we make use of the

standard assumption that the true (log) fundamental value of the stock price ¯ s

f

equals zero. Thus,

the stock price s

t

also equals the stock price misalignment.

11

The approach is, for example, also used in Westerhoﬀ and Dieci (2006) who model two ﬁnancial markets and their

interaction when introducing transaction taxes. Demary (2010) also analyzes the eﬀects of introducing such taxes in

a basic Westerhoﬀ-model augmented by diﬀerent time horizons of investors.

12

Negative orders denote a supply of stock.

13

There are also agent-based ﬁnancial models that make use of Walrasian market clearing. See for example Brock and

Hommes (1998).

5

Chartists expect that the direction of the recently observed price trend is going to continue:

˜

E

C

t

[s

t+1

− s

t

] = k

C

[s

t

− s

t−1

] (2)

k

C

is a positive parameter that denotes the strength of trend extrapolation. The tilde on the

expectations operator indicates that the expectation is not formed rationally. Fundamentalists, on

the other hand, expect that k

F

· 100 % of the actual perceived mispricing is corrected during the

next period:

˜

E

F

t

[s

t+1

− s

t

] = k

F

_

s

f

t

− s

t

_

(3)

s

f

t

is the perceived fundamental value that does not necessarily equal its true counterpart ¯ s

f

. The

diﬀerence between s

f

t

and ¯ s

f

is explained in detail in subsection 2.3. Assuming that the demand

generated by each type of investors depends positively on the expected price development leads to:

D

i

t

= ℓ E

i

t

[s

t+1

− s

t

] + ǫ

i

t

i = {C, F} (4)

ℓ is a positive reaction parameter. Since (2) and (3) do not reﬂect the great amount of chartist

and fundamentalist trading strategies that exist in real world markets, the noise term ǫ

i

t

is added.

It is normally distributed with standard deviation σ

i

and can be interpreted as the inﬂuence of

all other forecasting strategies diﬀerent from (2) and (3). The demand generated by chartist and

fundamentalist trading rules is therefore given by:

14

D

C

t

= b (s

t

− s

t−1

) + ǫ

C

t

b = ℓ · k

C

(5)

D

F

t

= c

_

s

f

t

− s

t

_

+ ǫ

F

t

c = ℓ · k

F

(6)

The fractions of agents using the two diﬀerent investment strategies are not ﬁxed over time.

Instead, agents continuously evaluate the strategies they use according to past performance. The

better a strategy performs relative to the other, the more likely it is that agents will employ it. It

14

Westerhoﬀ (2008) directly assumes eq. (5) and (6) and does not explicitly state the diﬀerent types of expectation

formations.

6

is assumed that the attractiveness of a particular strategy depends on its most recent performance

(exp{s

t

} − exp{s

t−1

}) D

i

t−2

as well as its past attractiveness A

i

t−1

:

15

A

i

t

= (exp{s

t

} − exp{s

t−1

}) D

i

t−2

+ dA

i

t−1

i = {C, F} (7)

The memory parameter 0 ≤ d ≤ 1 deﬁnes the strength with which agents discount past proﬁts.

The extreme cases d = 0 and d = 1 relate to scenarios where agents have zero and inﬁnite memory.

Note the timing of the model: Orders submitted in t − 2 are executed in t − 1. Their proﬁtability

ultimately depends on the price realization in t. Agents may also withdraw from trading (strategy

“0”). The attractiveness of this strategy A

0

t

is normalized to zero

_

A

0

t

= 0

_

. The fraction of agents

that employ strategy i is given by the well known discrete choice or Gibbs probabilities:

16

W

i

t

=

exp{eA

i

t

}

exp{eA

C

t

} + exp{eA

F

t

} + exp{eA

0

t

}

i = {C, F, 0} (8)

The more attractive a strategy, the higher the fraction of agents using it. Note that the probability

of choosing one of the three strategies is bounded between zero and one. The positive parameter e

measures the intensity of choice. The higher (lower) e, the greater (lesser) the fraction of agents that

will employ the strategy with the highest attractiveness. This parameter is often called the rationality

parameter in ABC ﬁnancial market models.

17

The described mechanism can be interpreted as an

evolutionary survival of the most proﬁtable forecasting strategy.

The only diﬀerence between our ﬁnancial market submodel and that of Westerhoﬀ (2008) is that

we distinguish between the true fundamental value ¯ s

f

and the trader’s perception of it, s

f

t

. Both

models are equivalent if s

f

t

= ¯ s

f

.

15

Recall that s

t

is the logarithm of the stock price. In order to calculate nominal proﬁts, s

t

has to be delogarithmized.

16

See, e.g., Manski and McFadden (1981) for a detailed explanation of discrete choice models.

17

Consult Westerhoﬀ and Dieci (2006), Hommes (2006) and Westerhoﬀ (2008).

7

2.2 Real Markets

The partial model describing the real sector is given by a hybrid NKM model. New Keynesian

models are widely used in macroeconomics because they typically allow for a good ﬁt of real world

data, and they are derived from individual optimization.

i

q

= δ

π

˜

E

q

[π

q+1

] + δ

x

˜

E

q

[x

q+1

] + ǫ

i

q

(9)

x

q

= χ

˜

E

q

[x

q+1

] + (1 − χ)x

q−1

−

1

σ

_

i

q

−

˜

E

q

[π

q+1

]

_

+ ǫ

x

q

(10)

π

q

= β

_

ψ

˜

E

q

[π

q+1

] + (1 − ψ)π

q−1

_

+ γx

q

− κs

q

+ ǫ

π

q

(11)

The notation of the variables is as follows: i is the deviation of the nominal interest rate from

its target, π the deviation of the inﬂation rate from its target, x the (log) output gap (i.e. its

deviation from steady state), and s the deviation of the (log) nominal stock price from its true

fundamental value ¯ s

f

. The subscript q = 1, ..., Q denotes the time index. We keep the common

interpretation of the time index in New Keynesian models and assume that it denotes quarters.

˜

E

q

[·] is the expectations operator conditional on knowledge available in q, where the tilde indicates

that they are formed non-rationally. The dynamic path of the stock price s is determined by the

model developed in the previous subsection. The variables ǫ

i

q

, ǫ

x

q

, ǫ

π

q

are stochastic elements with

zero mean.

Equation (9) is a standard monetary policy interest rule. The central bank reacts to expected

deviations of inﬂation and output from its target. For now, we use equal expectations formation

for the central bank and the market. In section 5.2 we will generalize the model by assuming that

the central bank’s expectations are diﬀerent from those of the market. Equation (10) is referred to

as the dynamic IS-curve that describes the demand side of the economy. It results from the Euler

equation (which is the result of intertemporal utility maximization) and market clearing in the goods

market. Equation (11) is a New Keynesian Phillips curve that represents the supply side. It can be

derived under the assumptions of nominal price rigidity and monopolistic competition. Asset prices

inﬂuence the economy through a balance sheet channel that works as follows: The willingness of

banks to grant credits typically depends on the borrowers’ ﬁnancial position. For example, agents

could use assets they hold as collateral when borrowing money. The more collateral a debtor has

to oﬀer, the more advantageous his credit contract will be. In this context, “advantageous” may

8

mean that either credits of larger size are oﬀered or that credits of the same size could be obtained

cheaper (lower interest payments). The ﬁrst argument can be used to relate asset prices positively to

aggregate demand, as for example done in Bernanke and Gertler (1999), Kontonikas and Ioannidis

(2005), Kontonikas and Montagnoli (2006), or Bask (2009). We stress the second argument in this

paper. Higher prices of assets owned by ﬁrms increase their creditworthiness, and allow them access

to cheaper credits. Since most ﬁrms’ production is largely ﬁnanced through credits, asset prices are

inversely related to ﬁrms marginal (real) costs of production. This argument allows the addition of

the term −κs

q

to equation (11).

18

This verbal kind of micro foundation is suﬃcient for our purposes.

The reader is referred to Bernanke and Gertler (1999) who discuss a balance sheet channel (and its

microfoundation) in more detail.

Note that we deﬁned s

q

as the nominal stock price gap. The so-called cost channel of monetary

transmission is commonly introduced into New Keynesian models by adding the nominal interest

rate into the Phillips-curve (see for example Ravenna and Walsh (2006) or Lam (2010)). Analogously

to this channel, we also decided to insert the nominal (and not the real) stock price gap into (11).

Note also that our deﬁnition of the stock price gap is very diﬀerent from that of Milani (2008) or

Castelnuovo and Nistico (2010), who deﬁne it as the diﬀerence between the stock price under fully

ﬂexible and somewhat rigid market conditions. Both, of course, are the result of utility optimal paths

under rational expectations. ABC ﬁnancial market models could also be employed for the analysis of

foreign exchange rates. Since a rise (fall) of foreign exchange rates would also raise (lower) production

costs – via more expensive (cheaper) intermediate inputs – they would be included with the opposite

sign (i.e. +κs

q

). To avoid confusion, we want to point out again that we are modeling stock prices

with the ABC submodel and not foreign exchange rates.

To derive eq. (10), it is commonly assumed that the household’s only possibility of transferring

wealth into future periods is by demanding bonds. Households therefore do not hold or trade stock.

We keep this assumption in order to allow for analysis of the isolated impact of the speculation of

ﬁnancial market participants on stock prices. We further assume that ﬁrms hold an initial amount of

stock but do not participate in stock trading. Consequently, they are only aﬀected by the ﬁnancial

sector via the balance sheet channel, and not via speculative gains. The ﬁnancial sector can not

generate proﬁts on the aggregate level by selling and reselling stock. If one agent wins from a

18

Formally, the cost channel is typically introduced by adding interest costs (+α · i

q

) into the Pillips-Curve. If interest

costs are negatively depending on solvency and thus stock prices, the term −κ · s

q

has to be added instead.

9

beneﬁcial transaction, others must lose. The only possibility for the aggregate stock market to

earn proﬁts is by dividend payments from the real sector. Because their relative size is small when

calculated for a daily basis, and because the Westerhoﬀ-model does not explicitly take ﬁnancial

wealth into account, we do not model the stream of dividend payments from ﬁrms to ﬁnancial

investors. As a result of the above arguments and assumptions, ﬁnancial streams between the real

and ﬁnancial sector do not exist.

The model can be rearranged as follows:

_

_

_

_

_

1 0 0

1

σ

1 0

0 −γ 1

_

_

_

_

_

. ¸¸ .

A

_

_

_

_

_

i

q

x

q

π

q

_

_

_

_

_

=

_

_

_

_

_

0 δ

x

δ

π

0 χ

1

σ

0 0 βψ

_

_

_

_

_

. ¸¸ .

B

_

_

_

_

_

0

˜

E

q

[x

q+1

]

˜

E

q

[π

q+1

]

_

_

_

_

_

+

_

_

_

_

_

0 0 0

0 1 − χ 0

0 0 β(1 − ψ)

_

_

_

_

_

. ¸¸ .

C

_

_

_

_

_

i

q−1

x

q−1

π

q−1

_

_

_

_

_

+

_

_

_

_

_

0

0

−κ

_

_

_

_

_

s

q

+

_

_

_

_

_

ǫ

i

q

ǫ

x

q

ǫ

π

q

_

_

_

_

_

(12)

The dynamics of the forward-looking variables i, x and π depend on the future expectations

˜

E

q

[x

q+1

],

˜

E

q

[π

q+1

]), their own history (i

q−1

, x

q−1

, π

q−1

), the current value of s

q

and the realizations of the

noise terms (ǫ

i

q

, ǫ

x

q

, ǫ

π

q

).

The endogenous real sector variables i

q

, x

q

and π

q

can be calculated as follows:

_

_

_

_

_

i

q

x

q

π

q

_

_

_

_

_

= A

−1

B

_

_

_

_

_

0

˜

E

q

[x

q+1

]

˜

E

q

[π

q+1

]

_

_

_

_

_

+A

−1

C

_

_

_

_

_

i

q−1

x

q−1

π

q−1

_

_

_

_

_

+A

−1

_

_

_

_

_

0

0

−κ

_

_

_

_

_

s

q

+A

−1

_

_

_

_

_

ǫ

i

q

ǫ

x

q

ǫ

π

q

_

_

_

_

_

(13)

Of course the parameters must be selected in a way that the system is stable. We take a closer

look on the stability conditions of the system in the next subsection after the model has been stated

completely.

Expectations in the real sector are also formed in a non-rational way. Following De Grauwe

(2010a) and De Grauwe (2010b) we assume that a certain fraction ω

opt

q

of agents is optimistic about

10

the future development of output while another ω

pes

q

is pessimistic. Both groups form expectations

according to:

Optimists expectation:

˜

E

opt

q

[x

q+1

] = g

t

(14)

Pessimists expectation:

˜

E

pes

q

[x

q+1

] = −g

t

_

= −

˜

E

opt

q

[x

q+1

]

_

(15)

The spread between the two expectations (2g

t

) is assumed to vary over time according to:

2g

t

= µ + ν · Std [x

t

] (16)

The parameters satisfy µ, ν ≥ 0 and Std [x

t

] denotes the unconditional standard deviation of the

output gap computed over a ﬁxed window of past observations.

19

The economic rationale behind this

implementation is that the agents beliefs diverge more when uncertainty surrounding the output gap

is high. In the special case of ν = 0 the divergence of beliefs is constant over time. In line with the

method provided in the previous subsection, we deﬁne the attractiveness of the diﬀerent strategies

as:

A

opt

q

= −

_

x

q−1

−

˜

E

opt

q−2

[x

q−1

]

_

2

+ ζA

opt

q−1

(17)

A

pes

q

= −

_

x

q−1

−

˜

E

pes

q−2

[x

q−1

]

_

2

+ ζA

pes

q−1

(18)

The attractiveness of forecasting strategies are therefore determined by past mean squared forecast

errors (MSFEs) weighted with decaying weights. Applying discrete choice theory, the weights that

determine the fractions of agents are given by:

ω

opt

q

=

exp{φA

opt

q

}

exp{φA

opt

q

} + exp{φA

pes

q

}

(19)

and ω

pes

q

=

exp{φA

pes

q

}

exp{φA

opt

q

} + exp{φA

pes

q

}

(20)

19

In all numerical simulations we set µ = 0.5 and ν = 2. The mentioned time windows is set to 20 periods.

11

The market’s expectation of the output gap is given by the weighted average of the two diﬀerent

forecasting strategies.

˜

E

q

[x

q+1

] = ω

opt

q

˜

E

opt

q

[x

q+1

] + ω

pes

q

˜

E

pes

q

[x

q+1

] (21)

Expectations about the inﬂation rate in De Grauwe (2010a) and De Grauwe (2010b) are formed

in a similar way. One type of agents (the targeters) believes in the inﬂation target that the central

bank has announced, hence their expectations are given by:

˜

E

tar

q

[π

q+1

] = π

⋆

(22)

Another group (the extrapolators) expect that the future inﬂation rate is given by the most recently

observed one, i.e. they extrapolate past values into the future. Expectations of this group are given

by:

˜

E

ext

q

[π

q+1

] = π

q−1

(23)

The markets expectation

˜

E

q

[π

q+1

] is again determined as the weighted average of these two groups.

Where the fractions of targeters and extrapolators (ω

tar

and ω

ext

) are again determined by the same

evolutionary approach used for expectations about the output gap.

Both expectations,

˜

E

q

[x

t+1

] and

˜

E

q

[π

t+1

], are not unrational. The diﬀerence to conventional

rational expectations (RE) is that no single agent is required to expect future dynamics rationally.

It has been pointed out by a number of authors, that forming conventional RE would indeed be

impossibly complicate.

20

It would require every agent to know how everybody else would react in

every possible situation and to calculate the resulting mean time paths in advance. It is implausible

that real world human beings are capable of solving such highly complex problems. In our model,

agents choose from a set of forecasting rules that are so simple that real world human beings would

be able to employ them. Using such simple rules is not unrational, it can be understood as the

best way to deal with an overwhelmingly complex world. An evolutionary mechanism is used to

permanently evaluate these strategies and sort out the poorly performing in favor of the better ones.

20

Consult Ackerman (2002), Gaﬀeo et al. (2008), Fair (2009) and Kirman (2010).

12

Hence, instead of requiring rationality from the individuals (as conventional rational expectations

do), it is the result of an evolutionary dynamic market process.

2.3 Bringing the Two Sectors Together

As already mentioned, the two parts of the model run on diﬀerent time scales. The real markets

operate quarterly while the ﬁnancial market operates daily. We assume that one quarter consists of

64 trading days. Therefore, the ﬁnancial sector performs 64 increments of the time index t within

one increment of the real market’s time index q (ﬁgure 1). Quarter q is deﬁned to contain the days

64(q − 1) + 1 , ... , 64q.

t: 1 2 3 ...

64 65 ... 128

q: 1 2

Figure 1: Time scale as indexed by days (t) and quarters (q)

We assume that the relevant value of the quarterly stock price s

q

that aﬀects the real sector via

eq. (13) is the average of the daily realizations of s

t

of the corresponding quarter q. Thus s

q

is given

by:

21

s

q

=

1

64

64q

t=64(q−1)+1

s

t

(24)

Using the deﬁnitions above, we calculate the recursive dynamics of the ﬁnancial market for one

quarter q (in days: t = (q − 1) · 64 + 1 , ... , q · 64) with the agent-based model deﬁned in section

2.1, and insert the mean of the resulting s

t

’s into eq. (13) in order to get the impact on real sector

variables.

Now that we have set up the real and ﬁnancial markets we can deﬁne the diﬀerence between the

true fundamental stock price (¯ s

f

) and the fundamentalist’s perception of it (s

f

t

). The fundamental

21

Eq. (24) assumes that the inﬂuence of daily stock prices on the real economy is equal for each day in the quarter. One

could instead also introduce a discounting factor into (24) to raise the relative inﬂuence of the more recent days. We

show in the online appendix that our results depend only marginally on the decision to use such a discounting factor.

13

Real

Markets

Financial

Market

Channel I: Cost EIIect

Channel II:

Misperception EIIect

Figure 2: Channels between real and ﬁnancial markets

value of any given stock is commonly understood to be the sum of all discounted future dividend

payments d

t+k

. In the most simple case it could be given by something similar to:

22

s

f

t

=

∞

k=1

ρ

k

E

t

[d

t+k

] (25)

Dividends are typically closely related to real economic conditions (x

q

in our model). Therefore, s

f

t

would depend on the expectation of x for all future days. We decided to model the perception of the

fundamental value in a diﬀerent way for two reasons: First, it has been empirically found that stock

markets overreact to new information, i.e. stock prices show stronger reactions to new information

than they should, given that agents behave rationally.

23

Second, it has been argued that in reality it

is very diﬃcult (if not impossible) to identify the true fundamental value of any stock.

24

Given these

problems, it seems reasonable to assume that agents do not know the true value of ¯ s

f

or calculate

it in a rational way (as in eq. (25)), but instead simply take the current development of the real

economy as a proxy for it.

s

f

t

= h · x

q

q = ﬂoor

_

t − 1

64

_

, h ≥ 0 (26)

The ﬂoor-function rounds a real number down to the next integer. Eq. (26) states that the funda-

mentalists’ perception s

f

t

is biased in the direction of the most recent real economic activity, i.e. if

output is high (low) the fundamental stock price is perceived to lie above (below) its true counter-

part. Note that ABC models of ﬁnancial markets can typically not relate the fundamental value

to the recent economic development, since the latter is not modeled endogenously. Most models do

22

Consult Campbell et al. (1997) chapter 7 for the derivation of this equation and more general versions.

23

De Bondt and Thaler (1985) were among the ﬁrst to describe this phenomenon.

24

For example Rudebusch (2005) or Bernanke and Gertler (1999) raise doubts of this kind.

14

0 0.5 1 1.5

0

0.2

0.4

0.6

0.8

1

h

κ

Stable Region

Unstable Region

(a) in h-κ-space

0 0.2 0.4 0.6 0.8 1

0

0.5

1

1.5

2

2.5

3

δ

x

δ

π

Stable Region

Unstable Region

(b) in δ

x

-δ

π

-space

Figure 3: Stability Analysis

26

not distinguish between s

f

t

and ¯ s

f

, they set both equal to zero or assume them to follow a random

walk.

25

Figure 2 illustrates the two channels that exist between the real and the ﬁnancial market.

Channel I (the cost channel) allows the ﬁnancial market to inﬂuence the real sector and disappears if

κ in eq. (11) is set equal to κ = 0. Channel II (the misperception of ¯ s

f

channel) allows for inﬂuence

in the opposite direction, and disappears if h in eq. (25) is set equal to h = 0. If both of these

cross-sectoral parameters are set equal to zero (κ = 0 & h = 0), both sectors (i.e. both submodels)

operate independently of each other.

The two cross-sectoral channels feed on each other. If stock prices are high, Channel I exerts a

positive inﬂuence on output: Solvency of ﬁrms rises which lowers their credit costs. Marginal costs

and thus inﬂation fall. As a result, output rises, which in turn exerts a positive inﬂuence on stock

prices through Channel II, and so on. To exclude explosive paths, κ has to be lower the higher h and

vice versa. Figure 3(a) shows a numerical approximation of the stability region in h-κ-space. It is

known that the policy parameters δ

π

and δ

x

are crucial for the stability of the NKM model. Under

standard speciﬁcation, a suﬃcient condition for stability is δ

π

> 1 and δ

x

≥ 0. To check whether

our behavioral model possesses a similar property, we generate a numerical approximation of the

stability region in δ

x

-δ

π

-space (ﬁgure 3(b)). The system is stable for δ

π

> 1 and δ

x

≥ 0 and thus

features the typical stability properties of NKM models.

25

Again, Westerhoﬀ (2008) is a good example to look at since both of these approaches are discussed there.

26

The parameterization used for this numerical investigation is discussed in detail below.

15

3 Numerical Simulations

The analysis of our model is performed by means of numerical simulation. The calibration is given in

Table 1. The parameter values for the ﬁnancial sector are exactly the same as in Westerhoﬀ (2008).

The values of σ, γ and β are so-called ”deep parameters” (or functions of such) and common in New

Keynesian models. For the policy parameters δ

x

and δ

π

we use the values that have originally been

suggested by Taylor (1993). The hybridity parameters χ and ψ are set to 0.8. These parameters

have to be set larger than 0.5 in order to maintain the endogenous business cycles of the De Grauwe

(2010a) model.

27

In order to set the cross-sectoral parameters, we assume that the real sector is

much less inﬂuenced by the ﬁnancial sector than the other way round.

28

Therefore we set h to be

ten times larger than κ.

Table 1: Baseline Calibration of the Model

Financial sector Real sector Interaction

a = 1 σ = 1 κ = 0.1

K

C

= 0.04 γ = 0.17166 h = 1

K

F

= 0.04 β = 0.99

ℓ = 1 δ

x

= 0.5

d = 0.975 δ

π

= 1.5

e = 300 ζ = 0.5

σ

s

= 0.01 φ = 10

σ

C

= 0.05 χ = 0.8

σ

F

= 0.01 ψ = 0.8

σ

ǫ

= 0.15

The memory parameter for the ﬁnancial sector d = 0.975 (which is taken from Westerhoﬀ (2008))

is much higher than that for the real sector ζ = 0.5 (taken from De Grauwe (2010a)). However,

when comparing these two values, it has to be taken into account that d refers to the discounting of

information that is only one day old. Bringing it to a quarterly basis we obtain d

64

= 0.2. Therefore

our calibration presumes that past information are less taken into account by ﬁnancial market agents

than by real sector agents. The intensity of choice parameters (e and φ) seam most problematic

27

One can easily see that if (9) is inserted into (10), the two terms containing

˜

E

q

[x

q+1

] cancel out if χ = 0.5. Whether

agents are optimistic or pessimistic does not play any role for the determination of x

q

any more. The self-fulﬁlling

character of expectations (explained in detail below) will ultimately break down.

28

We have assumed that the value of x is taken as information for the development of the real sector. If stock prices

overreact to new information (De Bondt and Thaler (1985), Nam et al. (2001), Becker et al. (2007)), this implies a

strong reaction of s to x and thus a high value of h.

16

since their impact on the results depends on the latent attractiveness values. We will therefore check

the robustness of our obtained result to diﬀerent parameterization of e and φ in section 5.

3.1 Dynamics of one Simulation

To demonstrate the working of our model, we perform one “representative” run. The simulated

time period consists of 100 quarters (or 6400 days). To eliminate the inﬂuence of arbitrary initial

conditions each simulation is performed with a ”burn-in” phase of 20 quarters. Figure 4 shows

the resulting dynamics for x

q

, π

q

, ω

opt

, ω

tar

, i

q

,

_

i

q

−

˜

E

q

[π

q+1

]

_

, s

t

, and a variable called market

sentiments. The latter represents the fraction of agents, employing the three trading strategies.

Black denotes chartist trading (W

C

), gray fundamentalist trading (W

F

), and white no trading

(W

0

). To generate the dynamics, a series of pseudo random numbers has to be drawn. Hence each

realization of simulated data is a unique result of the underlying random seed. The horizontal time

axes are quarterly scaled. In the diagrams containing daily data, quarters cover an interval of 64

data points.

The output gap is characterized by cyclical ups and downs. Hence the model generates an

endogenous business cycle. Closely connected to the up and down phases of output is the fraction of

optimists (ω

opt

). This result obviously follows from the speciﬁcation of our learning mechanism. In

times of high output, the forecasting rule of optimists (14) performs much better than the pessimists

rule (15). This results in a larger fraction of optimists and thus in a market expectation above

the steady state of zero

_

˜

E

q

[x

q+1

] > 0

_

. Since the resulting high output, again, favors optimist

forecasting, the situation has a tendency to reproduce itself. Vice versa for phases dominated by

pessimists. Hence, expectations about the output gap have a self-fulﬁlling nature and generate

business cycles.

A similar pattern can be observed for expectation formation of the inﬂation rate. However, the

self-fulﬁlling tendency is not as strong as for the output gap. If inﬂation largely deviates from the

target (e.g. around q = 35), the targeter’s expectation rule performs poorly relative to that of the

extrapolators. As a result, the fraction of targeters (ω

tar

) decreases and inﬂation can become very

low. In contrast to the expectations of the output gap, the mechanism is not strong enough to

generate a self-fulﬁlling, cyclical regime switching.

The stock market is also characterized by the emergence of diﬀerent regimes. Most of the time,

17

20 40 60 80 100

-0.5

0

0.5

O

u

t

p

u

t

G

a

p

20 40 60 80 100

-0.5

0

0.5

I

n

f

l

a

t

i

o

n

20 40 60 80 100

0

0.2

0.4

0.6

0.8

1

ω

o

p

t

20 40 60 80 100

0

0.2

0.4

0.6

0.8

1

ω

t

a

r

20 40 60 80 100

-1

-0.5

0

0.5

1

I

n

t

e

r

e

s

t

R

a

t

e

20 40 60 80 100

-0.5

0

0.5

R

e

a

l

I

n

e

t

e

r

s

t

R

a

t

e

10 20 30 40 50 60 70 80 90 100

-0.5

0

0.5

1

S

t

o

c

k

P

r

i

c

e

(

d

a

i

l

y

)

s

f

10 20 30 40 50 60 70 80 90 100

0

0.2

0.4

0.6

0.8

1

M

a

r

k

e

t

S

e

n

t

i

m

e

n

t

s

(

d

a

i

l

y

)

W

C

W

F

W

0

Figure 4: Model Output for a Time Period of Q = 40

18

when a certain amount of fundamentalists is present in the market, the stock price follows its

perceived fundamental value (which is given by the dashed line) closely. Deviations from s

f

during

such phases are transitory and small in size. The course of the stock price is thus largely inﬂuenced

by the underlying real economy: s

t

is high during the booms (q = 30, ..., 40 and q = 80, ..., 95) and

low during recessions (q = 10, ..., 20 and q = 59, ..., 70). If the market sentiments change in favor of

the chartists, stock prices become disconnected from the underlying fundamentals. Around q = 60,

for example, chartists form the dominating majority and the stock price moves away from s

f

. It

continues to follow a slight upward trend, although the underlying value falls, i.e. a bubble builds up.

In q = 62 market sentiments turn around, fundamentalists, who judge s

t

as extremely over-valued,

become dominating and drive the price down again, i.e. the bubble bursts. The opposite case of

chartists driving s

t

down below the fundamental value can be found for example around q = 45.

The model generates endogenous waves of optimism and pessimism, inﬂation-targeting and

inﬂation-extrapolation as well as chartism and fundamentalism. Each forecasting strategy is able to

dominate the market from time to time, but the evolutionary learning assures that none dominates

forever. The result is an endogenously occurring business cycle and endogenous stock price bubbles.

3.2 Impulse Response Analysis

In this subsection we analyze the eﬀects of an exogenous shock to the real sector. In DSGE models,

such questions are typically analyzed via impulse response functions that try to isolate the eﬀects

of an exogenous realization of the stochastic terms ǫ

i

q

, ǫ

x

q

and ǫ

π

q

. We focus on the impact of an

unanticipated, transitory cost shock without persistence of size ǫ

π,+

5

= 1. In order to allow for

impulse response analysis in a way similar to that typically used in DSGE models, we perform the

following experiment:

1. Generate the model dynamics for one particular random seed.

2. Generate the same dynamics with the same random seed (i.e. identical realizations of the

pseudo random numbers), but with ǫ

π

5

increased by ǫ

π,+

5

= 1.

3. Calculate the diﬀerences between the trajectories of step 1 and 2 which gives the isolated

impact of the cost shock. Note that the noise terms are identical in both runs. Diﬀerences are

thus not a result of diﬀerent random numbers, but solely due to the imposed shock.

4. Repeat steps 1-3 10,000 times.

19

0 10 20 30

−1

−0.5

0

O

u

t

p

u

t

G

a

p

0 10 20 30

−1

−0.5

0

S

t

o

c

k

P

r

i

c

e

0 10 20 30

0

0.5

1

I

n

f

l

a

t

i

o

n

0 10 20 30

−0.5

0

0.5

R

e

a

l

I

n

t

e

r

e

s

t

R

a

t

e

Figure 5: Mean response to a exogenous cost shock of size one. Dashed lines are 95 % quantiles.

Figure 5 shows the resulting responses to an exogenous shock of ǫ

π,+

5

= 1 for our baseline cali-

bration. The solid lines illustrate mean responses, while the dashed lines represent 95% quantiles.

On average, the economy shows the typical stagﬂationary response to the cost shock. Inﬂation and

the real interest rate rise, while output and the stock price fall. All impulse responses show high

volatility. The quantiles for the output gap, for example, illustrate that the reaction of x

q

can be

located anywhere between (a) a strong negative reaction on impact that is accelerated during the

subsequent two periods and followed by a hump-shaped path back towards trend and (b) no reaction

on impact followed by a slightly positive path in the medium and long run. The other time series

exhibit similarly volatile impulse responses. The only exception is (by construction of the shock)

the reaction of inﬂation on impact.

To analyze the source of this high volatility, we generate the impulse response functions of π and

s that result if either optimists or pessimists are dominating the market during the shock period. We

deﬁne a situation in which ω

opt

q

≥ 0.75 as dominated by optimists and a situation in which ω

pes

q

≥ 0.75

as dominated by pessimists. The top row in ﬁgure 6 shows the eﬀect on inﬂation and stock prices.

The reaction of stock prices is stronger when optimists dominate. If pessimists form the majority,

the amplitude is smaller for both time series. The economic logic underlying this phenomenon is

the following. If the number of optimists is high, the economy is caught in a self-reproducing circle

of high output and optimistic expectations. A huge contractionary shock can break this circle and

thus turn the boom into a recession. If pessimists dominate, such an ampliﬁcation mechanism can

20

0 10 20 30

0

0.5

1

Shock period dominated by ...

I

n

f

l

a

t

i

o

n

0 10 20 30

−0.2

−0.1

0

S

t

o

c

k

P

r

i

c

e

s

(

d

a

i

l

y

)

Optimists

Pessimists

0 10 20 30

0

0.5

1

Shock period dominated by ...

I

n

f

l

a

t

i

o

n

0 10 20 30

−0.4

−0.2

0

S

t

o

c

k

P

r

i

c

e

s

(

d

a

i

l

y

)

Targeters

Extrapolators

0 10 20 30

0

0.5

1

Shock period dominated by ...

I

n

f

l

a

t

i

o

n

0 10 20 30

−0.4

−0.3

−0.2

−0.1

0

S

t

o

c

k

P

r

i

c

e

s

(

d

a

i

l

y

)

Fundamentalist

Chartists

Figure 6: Mean response of output and stock price with initial conditions dominated by diﬀerent

groups of agents

not emerge and the fall in GDP is much smaller. The same contractionary shock therefore has a

higher mean impact during a boom than during a recession.

Comparing impulse responses for those cases when targeters dominate with those dominated by

extrapolators,

29

yields a similar picture (second row in ﬁgure 6). If the latter are dominating, the

impact of the shock is stronger on impact and more persistent for both time series. The economic

rationale is that extrapolators do not generate a mean-reversion. Since they expect the current

state of inﬂation to persist, they create persistence in the system. Persistence, in turn, makes their

own forecasting strategy more attractive and the number of extrapolators might increase further.

Therefore, the higher persistence in this case is also a result of self-fulﬁlling expectations. If targeters

dominate (i.e. the inﬂation target of the central bank is credible), the persistence of the shock is

much lower. Strong believes in a stable system obviously lead to a dampening of shocks. Whether

fundamentalists or chartists dominate the stock market does not play a major role in the transmission

of cost shocks.

29

Dominance has been deﬁned analogously to the above case.

21

The high volatility that has been found in the impulse responses of ﬁgure 5 is therefore partly a

result of the history dependence. Since the shock can have a diﬀerent impact on average depending

on the initial beliefs of agents, the uncertainty about the impact of the shock increases. De Grauwe

(2010a) has also analyzed the origin of persistence in his behavioral NKM and ﬁnds that responses

maintain persistent even if the hybrid character is turned oﬀ. This point is of interest, because

persistence has been a matter of concern in NKM modeling. In its baseline notation (for χ = 1

and ψ = 1 in (10)-(11)) those models do not produce persistent responses to non-persistent shocks.

De Grauwe argues that the evolutionary learning algorithm produces endogenous persistence, while

in standard NKM models persistence is introduced exogenously by assuming a hybrid form. Now

that we have gained some understanding of the dynamics of the model, we can use it to analyze a

prevailing question currently debated among policy makers.

4 Taxing Financial Transactions

The recent ﬁnancial crisis has created enormous costs in all industrialized economies. First, it

produced a huge decline in GDP and rise in unemployment. Second, the ﬁnancial positions of

states have been very negatively aﬀected because of several necessary stabilization policies like

capital injections, purchase of assets, ﬁscal stimuli, direct support and many more. On average, the

advanced G20 countries suﬀered a rise in government debt by 40%.

30

Because of such rising debt,

several countries directly stumbled from the ﬁnancial- into the ﬁscal crisis. As a response to those

devastating externalities of the ﬁnancial sector, it should be asked, ﬁrst, whether new regulatory

policies are needed to stabilize this sector for the future and, second, whether it should provide a

ﬁnancial contribution to the recently generated costs. The traditional way of achieving both would

be to levy a ﬁnancial transaction tax (FTT) in the spirit of Tobin (1978). Such a tax would make

short term trading less attractive, while having no signiﬁcant inﬂuence on long term trading. Since

high frequency, speculative trading is a socially wasteful business, it would be beneﬁcial to curtail it

by introducing a FTT. Long run oriented trading that is based on underlying fundamental values,

is not aﬀected.

31

In approaching this question, the G20 leaders have recently asked the IMF to prepare a report

30

Consult IMF (2010).

31

A nice executive summary of arguments in favor and against a FTT can be found in Schulmeister et al. (2008).

22

on how the ”ﬁnancial sector could make a fair and substantial contribution” in bearing parts of

the induced burden.

32

Summarizing the results of this report, the IMF argues that taxing ﬁnancial

transaction is generally a feasible policy instrument for achieving this goal and that ”the FTT should

not be dismissed on grounds of administrative practicality” (p. 19). However it also argues that the

traditional FTT might not be the best instrument to ”ﬁnance a resolution mechanism” and ”focus

on core sources of ﬁnancial instability”. Another slightly diﬀerent type of tax – called ﬁnancial

activities tax (FAT) – might be better suited than the FTT.

While policy makers are currently intensively debating the introduction of such taxes,

33

one

striking aspect of the debate is ”that it is almost entirely unguided by the public ﬁnance literature

on the topic – because there is hardly any”

34

. In this section, we use our model to analyze the

advantages and disadvantages of both kinds of taxes with regard to their ability to stabilize markets

and to raise ﬁscal income. To evaluate their eﬀect on the variables of interest, we report the average

fractions of fundamentalists

1

T

T

t=1

W

F

t

and chartists

1

T

T

t=1

W

C

t

resulting from a certain tax. We

also report the average tax revenue per agent and day that is given as:

Tax Revenue =

1

T

T

t=1

_

W

F

t

· Tax

F

t

+ W

C

t

· Tax

C

t

_

(27)

Where Tax

F

is the tax payed by fundamentalists and Tax

C

the tax payed by chartists. Additionally

we deﬁne the following two measures:

35

vol(s) =

1

T − 1

T

t=2

|s

t−1

− s

t

| dis(s) =

1

T

T

t=1

|s

t

| (28)

And for quarterly time series:

vol(z) =

1

Q − 1

Q

q=2

|z

q−1

− z

q

| dis(z) =

1

Q

Q

q=1

|z

q

| z = {x, π} (29)

The measure vol(·) denotes the volatility (i.e. rate of change) of a time series. Accordingly, dis(·)

measures its distortion (i.e. diﬀerence to fundamental steady state). We do not use the variance

32

The mentioned report is IMF (2010).

33

Besides the IMF and the G20, the European Commission and the European Parliament are currently examining

weather a ﬁnancial tax should be introduced. See for example EU (2010), EU (2011a) or EU (2011b).

34

Quote taken from Keen (2011).

35

Both measures closely follow Westerhoﬀ (2008).

23

measure because it interprets volatility via the average squared distance from the mean. Our time

series show long-lasting deviations from the mean (which we interpret as bubbles or distortion).

When calculating the variance, one would not measure the volatility but rather the mean squared

distortion. To avoid confusion we do not use the variance measure.

The introduction of a FTT has already often been analyzed in the ABC ﬁnance literature. Ex-

amples are Westerhoﬀ (2003), Westerhoﬀ (2008) or Demary (2008). These studies, however, limit

their attention to the reduction of volatility and distortion of stock prices. Our study adds several

aspects to this literature: (1) We do not restrict our analysis to the stabilization of ﬁnancial mar-

kets but include real markets as well. (2) We contrast the classical FTT with the innovative, very

recently proposed FAT. (3) We also answer a question that has become very prevailing during the

recent ﬁscal crisis: how should a tax be designed in order to yield maximal tax revenues? There

has also been a number of empirical studies that investigate the impact of FTTs.

36

These studies,

however, focus on short-term volatility and neglect long-term mispricing. We take the latter into

account, since it can lead to the built up and bursting of bubbles and therefore might have the most

important impact on the real economy.

4.1 Financial Transaction Tax

The basic characteristics of the FTT is that it is small in size but levied on a brought basis: the total

value of transaction. To introduce it into our model, we assume that the tax has to be paid relative

to the nominal value traded. Since complete investment consists of two transactions, the tax also has

to be paid twice. Orders generated in D

t−2

imply nominal transactions of D

t−2

· exp{s

t−1

} in t − 1

and D

t−2

· exp{s

t

} in t. The tax rate τ is applied to the absolute nominal value of both transactions

(i.e. buys and sells are equally taxed). Since tax payments directly reduce the proﬁtability of an

investment, eq. (7) changes to:

37

A

i

t

= (exp{s

t

} − exp{s

t−1

}) D

i

t−2

− τ (exp{s

t

} + exp{s

t−1

})

¸

¸

D

i

t−2

¸

¸

+ dA

i

t−1

(30)

The transaction tax is represented by τ and

¸

¸

D

i

t−2

¸

¸

is the absolute value of D

i

t−2

. We run the model

for 500 quarters (32,000 days) with diﬀerent values for τ as well as 1000 diﬀerent realizations of the

36

See IMF (2010) p. 20 for a summary of those studies.

37

Consult also Westerhoﬀ (2008) for the introduction of an FTT into his model.

24

0 0.5 1

0.005

0.01

0.015

0.02

v

o

l

(

s

t

o

c

k

p

r

i

c

e

)

Tax Rate

0 0.5 1

0.2

0.3

0.4

0.5

d

i

s

(

s

t

o

c

k

p

r

i

c

e

)

Tax Rate

0 0.5 1

0.21

0.212

0.214

0.216

0.218

0.22

v

o

l

(

o

u

t

p

u

t

g

a

p

)

Tax Rate

0 0.5 1

0.24

0.25

0.26

0.27

0.28

d

i

s

(

o

u

t

p

u

t

g

a

p

)

Tax Rate

0 0.5 1

0.135

0.136

0.137

0.138

0.139

0.14

v

o

l

(

i

n

f

l

a

t

i

o

n

)

Tax Rate

0 0.5 1

0.16

0.18

0.2

0.22

0.24

d

i

s

(

i

n

f

l

a

t

i

o

n

)

Tax Rate

0 0.5 1

0

0.1

0.2

0.3

0.4

0.5

f

u

n

d

a

m

e

n

t

a

l

i

s

t

s

(

s

o

l

i

d

)

c

h

a

r

t

i

s

t

s

(

d

a

s

h

e

d

)

Tax Rate

0 0.5 1

0

1

2

3

4

x 10

-5

T

a

x

R

e

v

e

n

u

e

Tax Rate

Figure 7: Impact of Financial Transaction Tax (FTT)

pseudo random number generator for each τ. Figure 7 shows the average fraction of chartists and

fundamentalists, the gained tax revenue as well as volatility and distortion of s, x and π with respect

to the imposed FTT.

Increasing the tax rate (starting from zero), leads to a sharp decline in the fraction of chartist

traders which approximately equals zero for τ ≥ 0.6%. At the same time it slightly increases the

number of fundamentalist traders up to τ ≈ 0.1% and decreases it gradually for higher tax rates.

Tax revenue follows a typical Laﬀer curve: Increasing the tax rate up to τ ≈ 0.23% leads to rising

tax revenue. But increasing the tax rate further crowds too many agents out of the market and thus

leads to a falling tax income.

Concerning stability, we evaluate the FTT by how well it is capable of reducing volatility and

distortion of s, x and π. With respect to vol(s) and vol(π), the FTT has an exclusively positive

inﬂuence on stability. Increasing the FTT leads to monotonically decreasing volatility of stock prices

and inﬂation. The measures vol(x), dis(s), dis(x) and dis(π) recommend a diﬀerent conclusion. All

four follow a u-shaped pattern with minimum near τ = 0.3%. Therefore, with respect to these

variables, the FTT has an ambiguous impact. It stabilizes the market for small tax rates. If it

becomes too large (τ > 0.3%) the market is destabilized and the values of dis(s), dis(x) and dis(π)

become quickly very large.

25

4.2 Financial Activities Tax

A FAT, as proposed in the report of the IMF (2010) is ”levied on the sum of proﬁts and remuneration

of ﬁnancial institutions” (p. 21). Several detailed examples of how such a FAT could look like

can be found in the report (p. 66-70). In our model, we have to use a more stylized version of

course. Since, we do not consider labor costs in the ﬁnancial sector, all gains from stock trading

(exp{s

t

} − exp{s

t−1

}) D

i

t−2

are proﬁts. Thus we introduce a FAT into our model by taxing proﬁts

a constant rate of τ. If we assume further that the FAT only applies if proﬁts are positive, tax

payments are given by:

1

{R

+

}

_

(exp{s

t

} − exp{s

t−1

}) D

i

t−2

¸

· τ · (exp{s

t

} − exp{s

t−1

}) D

i

t−2

Where 1

{R

+

}

[y] is the indicator function that becomes 1 if y ∈ R

+

and zero otherwise. Equation

(7) changes to:

A

i

t

=(exp{s

t

} − exp{s

t−1

}) D

i

t−2

+ dA

i

t−1

(31)

−1

{R

+

}

_

(exp{s

t

} − exp{s

t−1

}) D

i

t−2

¸

· τ · (exp{s

t

} − exp{s

t−1

}) D

i

t−2

(32)

We perform the same experiment that we used to analyze the impact of the FTT. Results are

illustrated in ﬁgure 8. The rate of a FAT has to be much higher than that of a FTT because its

base is much smaller. In our analysis we account for tax rates between 0% and 50%. To allow for a

better comparability of the results, scaling of the ordinates is carried over from ﬁgure 7.

The tax slowly decreases the number of chartist while keeping the number of fundamentalists

almost constant. All other measures follow a monotonic path. Tax revenues rise while volatility

and distortion of s, x and π fall. The eﬀect of the FAT is thus clearly positive. Increasing it leads

to an improvement of all of our measures. An optimal value of τ can not be identiﬁed. Therefore

the question for the FAT’s size is not a question of economic optimality, but mainly one of political

feasibility.

For comparing both taxes with each other, we assume that the FTT is set in the range of optimal

values at τ = 0.28%. This rate leads to low market volatility and distortion at a high revenue

(section 4.1). For the FAT we assume diﬀerent values between 5% and 50%.

38

The impact on all

38

For illustration purpose, IMF (2010) [p. 22, 69] and EU (2010) [p. 6] assume a rate of 5%.

26

0 20 40

0.005

0.01

0.015

0.02

v

o

l

(

s

t

o

c

k

p

r

i

c

e

)

Tax Rate

0 20 40

0.2

0.3

0.4

0.5

d

i

s

(

s

t

o

c

k

p

r

i

c

e

)

Tax Rate

0 20 40

0.21

0.212

0.214

0.216

0.218

0.22

v

o

l

(

o

u

t

p

u

t

g

a

p

)

Tax Rate

0 20 40

0.24

0.25

0.26

0.27

0.28

d

i

s

(

o

u

t

p

u

t

g

a

p

)

Tax Rate

0 20 40

0.135

0.136

0.137

0.138

0.139

0.14

v

o

l

(

i

n

f

l

a

t

i

o

n

)

Tax Rate

0 20 40

0.16

0.18

0.2

0.22

0.24

d

i

s

(

i

n

f

l

a

t

i

o

n

)

Tax Rate

0 20 40

0

0.1

0.2

0.3

0.4

0.5

f

u

n

d

a

m

e

n

t

a

l

i

s

t

s

(

s

o

l

i

d

)

c

h

a

r

t

i

s

t

s

(

d

a

s

h

e

d

)

Tax Rate

0 20 40

0

1

2

3

4

x 10

-5

T

a

x

R

e

v

e

n

u

e

Tax Rate

Figure 8: Impact of Financial Activities Tax (FAT)

measures of interest are contrasted in table 2. For all values, the FAT leads to a higher fraction of

fundamentalists. All other measures are in favor of the FTT. The tax revenue is larger or at least

equally good for the FTT while all undesirable characteristics (W

C

, dis(·) and vol(·)) are smaller.

Note however, that a large fraction of fundamentalists is not an end in itself, it is only a useful

means to stabilize the market. Therefore, it is only a measure of second order compared to values

of distortion and volatility. Since the FTT leads to better (or at least equally good) results in all

other respect, we can conclude that the FTT strictly dominates the FAT.

A point in favor of the FAT is that there is no danger of setting the rate too high. If the optimal

rate for the FTT is missed and a larger rate is set (0.8% for example), huge distortions might occur

(ﬁgure 7). The FAT does not suﬀer from such a problem (ﬁgure 8) because of its monotonic impact.

The IMF also considers another variant of the FAT in its report and proposes: ”taxing high

returns more heavily than low” (p. 68). Adding such an element of progressivity, should discourage

risk-taking. To test the scope of such a progressive FAT (called FAT3 in the IMF report), we

perform the above experiment a third time. Instead of assuming a ﬂat tax as in equation (31) we

let the tax rate grow with proﬁts. The ﬁrst step in deﬁning such a tax, is to identify a threshold

value of proﬁts. Proﬁts above this threshold are deﬁned ’excess proﬁts’ and thus taxed higher. Let

27

Table 2: Comparison of FTT with FAT

FTT FAT

0.28% 5% 10% 20% 35% 50%

W

C

0.054 0.297 0.273 0.231 0.179 0.142

W

F

0.367 0.400

a

0.402 0.404 0.403 0.399

Revenue 3.2 ·10

−5

0.9 ·10

−5

1.6 ·10

−5

2.5 ·10

−5

2.9 ·10

−5

3.0 ·10

−5

dis (s) 0.228 0.256 0.251 0.243 0.234 0.229

dis (x) 0.249 0.250 0.250 0.250 0.249 0.249

dis (π) 0.171 0.174 0.173 0.173 0.172 0.171

vol (s) 0.009 0.016 0.015 0.014 0.012 0.011

vol (x) 0.211 0.212 0.212 0.212 0.212 0.211

vol (π) 0.139 0.140 0.139 0.139 0.139 0.139

a

Bold numbers indicate that the FAT leads to better results than the FTT

P

t

= (exp{s

t

} − exp{s

t−1

}) D

i

t−2

denote proﬁts. We deﬁne the benchmark P

⋆

to be the standard

deviation of proﬁts:

P

⋆

:= std(P

t

) (33)

A tax rate that is quadratically growing in P can be deﬁned as:

FAT rate = 1

{R

+

}

[P] · τ ·

_

P

P

⋆

_

2

(34)

Such a tax deﬁnition has the following nice properties. If proﬁts equal P

⋆

(the benchmark value),

they are taxed by a rate of τ (ﬁgure 9). If they are above P

⋆

(excess proﬁts) they are taxed by

a higher rate that grows quadratically in P.

39

For proﬁts below the threshold, the tax falls and

smoothly approaches zero at P = 0. The tax rate is thus progressive and not subject to any steps.

It also allows us to perform experiments similar to the ones above. By increasing (decreasing) τ, we

increase (decrease) the FAT for all (positive) values of P in the same direction while preserving the

general shape.

Figure 10 compares the results for the progressive FAT (bold line) with the ﬂat tax scenario of the

39

For very high proﬁts eq. (34) might result in tax rates above 100%. Since such rates are unrealistic, we restrict the

tax to values ≤ 100%.

28

0

0

Profits

F

A

T

r

a

t

e

P*

τ

Figure 9: Rate of the progressive FAT

previous experiment (thin line). Comparing both FAT’s leads to ambitious results. The progressive

FAT gives rise to a better stabilization for rates up to τ ≈ 30%: The number of fundamentalists

is higher while the number of chartists is lower compared to the ﬂat FAT. At the same time, all

volatility and distortion measures are lower. Only for high tax rates of 30% or more the results turn

around partly: vol(s), dis(s), vol(x), dis(x) and dis(π) are advantages under a ﬂat tax. The revenue

is larger under a progressive FAT between 0% and 11%. For tax rates above 11% the ﬂat FAT yields

more income. The nice property of monotonically decaying volatility and distortion that has already

been found for the ﬂat FAT is also found for the progressive one.

0 20 40

0.01

0.012

0.014

0.016

0.018

0.02

v

o

l

(

s

t

o

c

k

p

r

i

c

e

)

Tax Rate

0 20 40

0.22

0.23

0.24

0.25

0.26

0.27

d

i

s

(

s

t

o

c

k

p

r

i

c

e

)

Tax Rate

0 20 40

0.211

0.2112

0.2114

0.2116

v

o

l

(

o

u

t

p

u

t

g

a

p

)

Tax Rate

0 20 40

0.248

0.249

0.25

0.251

d

i

s

(

o

u

t

p

u

t

g

a

p

)

Tax Rate

0 20 40

0.139

0.1392

0.1394

0.1396

0.1398

0.14

v

o

l

(

i

n

f

l

a

t

i

o

n

)

Tax Rate

0 20 40

0.17

0.171

0.172

0.173

0.174

0.175

d

i

s

(

i

n

f

l

a

t

i

o

n

)

Tax Rate

0 20 40

0.1

0.2

0.3

0.4

0.5

f

u

n

d

a

m

e

n

t

a

l

i

s

t

s

(

s

o

l

i

d

)

c

h

a

r

t

i

s

t

s

(

d

a

s

h

e

d

)

Tax Rate

0 20 40

0

1

2

3

x 10

-5

T

a

x

R

e

v

e

n

u

e

Tax Rate

flat

progressive

Figure 10: Impact of progressive Financial Activities Tax (FAT)

The reason for this result can easily be explained by taking a look at the distribution of proﬁts.

Figure 11 shows the distribution of fundamentalist’s proﬁts (solid line) and chartist’s proﬁts (dashed

29

10

0

10

−1

10

−2

10

−3

10

−4

0

500

1000

1500

2000

Profits

P

r

o

b

a

b

i

l

i

t

y

Fundamentalists

Chartists

10

0

10

−1

10

−2

10

−3

10

−4

10

−4

10

−2

10

0

10

2

10

4

Profits

P

r

o

b

a

b

i

l

i

t

y

Fundamentalists

Chartists

Figure 11: Distribution of Proﬁts in a semi-log (left) and log-log scaled plot (right).

line), both, in a semi-log scaled plot and a log-log scaled plot. Both lines show the characteristic

fat tails.

40

At the same time, the distribution of fundamentalist’s proﬁts has more density located

at low values while chartists have more density located at higher values. In the tail (i.e. for very

high proﬁts) both distributions are approximately equal. Fundamentalists (who earn lower proﬁts

on average) are thus favored by the progressive FAT, while chartists (who earn higher proﬁts on

average) are disadvantaged. As a result, the fraction of fundamentalists increases while that of

chartists decreases compared to a ﬂat FAT (ﬁgure 10, upper left panel). A more stable economy is

the result.

Finally, we compare the progressive FAT at diﬀerent rates with the FTT of 0.28%. Results are

given in table 3. For all values of the tax, the fraction of fundamentalists is higher than under the

FTT while the values of vol (π) are equal throughout all simulations. If we leave the fractions of

fundamentalists and chartists aside, the FTT again strictly dominates the FAT. We can therefore

conclude that the classical FTT is better suited than the new FAT in order to achive stabilization

of markets and raise funds from the ﬁnancial sector.

40

Consult Lux (2009) on the empirical properties of ﬁnancial data.

30

Table 3: Comparison of FTT with progressive FAT

FTT FAT (progressive)

0.28% 5% 10% 20% 35% 50%

W

C

0.054 0.258 0.233 0.203 0.178 0.162

W

F

0.367 0.424

a

0.436 0.449 0.460 0.466

Revenue 3.2 ·10

−5

1.8 ·10

−5

1.7 ·10

−5

1.7 ·10

−5

1.6 ·10

−5

1.6 ·10

−5

dis (s) 0.228 0.243 0.239 0.235 0.234 0.233

dis (x) 0.249 0.250 0.249 0.249 0.249 0.249

dis (π) 0.171 0.172 0.172 0.172 0.172 0.171

vol (s) 0.009 0.015 0.014 0.013 0.012 0.012

vol (x) 0.211 0.212 0.212 0.212 0.212 0.212

vol (π) 0.139 0.139 0.139 0.139 0.139 0.139

a

Bold numbers indicate that the FAT leads to better results than the FTT

5 Robustnes Checks

In this section, we check the robustness of our result with respect to some assumptions that we had

to make throughout our analysis. Since our simulations suggests that the FTT is the best way of

taxing the ﬁnancial sector, we focus on the robustness of our derived optimal FTT.

5.1 Parameterization

As mentioned in section 3, calibration of the intensity of choice parameters φ and e is probably the

most problematic one. Our ﬁrst robustness checks are therefore concerned with these parameters.

Table 4 and 5 show how some important results change with the variation of φ and e.

Table 4: Robustness check of intensity of choice parameter φ

φ = 5 φ = 8 φ = 10 φ = 12 φ = 15

min

τ

vol(x) 0.205 0.209 0.211 0.212 0.214

arg min

τ

vol(x) 0.30% 0.28% 0.28% 0.28% 0.25%

min

τ

dis(x) 0.226 0.241 0.249 0.255 0.262

arg min

τ

dis(x) 0.30% 0.29% 0.28% 0.28% 0.27%

min

τ

Revenue 3.2 ·10

−5

3.2 ·10

−5

3.2 ·10

−5

3.2 ·10

−5

3.2 ·10

−5

arg min

τ

Revenue 0.26% 0.26% 0.26% 0.26% 0.26%

31

Our results are fairly robust against diﬀerent values of the parameter φ (table 4). The optimal

tax rate as well as the induced optimal values of vol(x), dis(x) and Revenue change only slightly.

The opposite holds for the parameter e. The optimal tax rate diverges strongly between 0.18%

and 0.54% for diﬀerent values of e (table 5). The induced Tax revenues are also subject to high

uncertainty.

This robustness test suggests, that reliable estimations of the intensity of choice parameter for the

ﬁnancial sector are of major importance for reliable policy suggestions. We have shown in section 4.1

that a too high FTT can result in huge distortion of the real economy. The negative eﬀects of a too

low FTT will instead be much smaller. A good strategy might thus be to set the FTT signiﬁcantly

below the value that is optimal with respect to a given parameterization in order to deal with the

uncertainty in an appropriately careful way.

Table 5: Robustness check of intensity of choice parameter e

e = 100 e = 200 e = 300 e = 400 e = 500

min

τ

vol(x) 0.211 0.211 0.211 0.211 0.211

arg min

τ

vol(x) 0.54% 0.36% 0.28% 0.25% 0.18%

min

τ

dis(x) 0.248 0.248 0.249 0.250 0.251

arg min

τ

dis(x) 0.80% 0.41% 0.28% 0.10% 0.12%

min

τ

Revenue 8.4 ·10

−5

4.4 ·10

−5

3.2 ·10

−5

2.7 ·10

−5

2.3 ·10

−5

arg min

τ

Revenue 0.81% 0.41% 0.26% 0.18% 0.16%

5.2 Taylor Rule

In this section we test the robustness against diﬀerent speciﬁcations of the Taylor rule. We begin

by introducing an interest smoothing parameter into (9). The current interest rate is thus not only

concerned with reducing inﬂation and the output gap but also with producing a smooth path of i

q

over time. Variation A of the policy rule is thus given by:

i

q

= λ

_

δ

π

˜

E

q

[π

q+1

] + δ

x

˜

E

q

[x

q+1

]

_

+ (1 − λ)i

q−1

+ ǫ

i

q

(35)

As a second variation, we assume that the board of the central bank is composed of a set of

heterogeneous agents who form expectations diﬀerently. Some are optimistic about future output

and some are pessimistic. The expectations and fractions of optimistic and pessimistic central

32

bankers are the same as for the other agents. Thus, they are given by

˜

E

opt

q

[x

q+1

],

˜

E

pes

q

[x

q+1

], ω

opt

q

and ω

pes

q

as deﬁned in section 2.2. Assume that the majority of central bankers can fully introduce

their expectations into the policy rule. The expectation that enters the Taylor rule is then given by:

ˆ

E

q

[x

q+1

] =

_

¸

_

¸

_

˜

E

opt

q

[x

q+1

] if ω

opt

q

> ω

pes

q

˜

E

pes

q

[x

q+1

] otherwise

(36)

In other words, the central bank’s decision structure is such that it become fully optimistic if optimists

dominate and fully pessimistic if pessimists dominate. Variation B of the Taylor rule is given by:

i

q

= δ

π

ˆ

E

q

[π

q+1

] + δ

x

ˆ

E

q

[x

q+1

] + ǫ

i

q

(37)

Where

ˆ

E

q

[π

q+1

] is formed analogous to

ˆ

E

q

[x

q+1

].

Table 6: Robustness check of diﬀerent versions of the Taylor rule

Taylor Rule

Baseline

speciﬁcation

Variation A

(λ = 0.75)

Variation B

min

τ

vol(x) 0.211 0.203 0.228

arg min

τ

vol(x) 0.28% 0.24% 0.20%

min

τ

dis(x) 0.249 0.278 0.272

arg min

τ

dis(x) 0.28% 0.24% 0.29%

min

τ

Revenue 3.2 ·10

−5

3.2 ·10

−5

3.3 ·10

−5

arg min

τ

Revenue 0.26% 0.26% 0.27%

Table 6 compares our result for the diﬀerent speciﬁcations of the Taylor rule. The parameter λ

in Variation A is set to λ = 0.75.

41

The results show that the measures of stability and tax revenue

are only subject to minor change. The optimal tax rate that minimizes vol(x) undergoes the largest

change. It falls by 0.04% in case of variation A and 0.08% for variation B. This uncertainty again

suggests that the rate of the FTT should be set to lower rates compared to those that are optimal

given our baseline speciﬁcation.

41

The same analysis for a wider λ-range as well as the impulse response functions resulting from the variation of the

Taylor Rules can be found in the online appendix.

33

6 Conclusion

We have developed a model that combines agent-based ﬁnancial market theory with New Keynesian

macroeconomics. The two employed submodels are simple representatives of their respective disci-

pline. They are both subject to an evolutionary process of expectation formation that sorts out the

poorly performing strategies in favor of the good ones. Interaction between the two models is brought

about by two straightforward channels. Our comprehensive model is very stylized and not yet ready

for econometric analysis. But even with this simplistic methodology, we are able to show that the

behavioral structure of our model has a strong impact on the transmission of shocks. The market

sentiments in the shock period, for example, can lead to a very diﬀerent average transmissions.

We also used the model to analyze a question that is currently debated among policy makers.

Namely, if the introduction of a tax on ﬁnancial transactions can bring about positive developments

for the overall economy. We ﬁnd that such a tax could generally reduce volatility and distortion of

the real and ﬁnancial market variables, but that its size and type plays an important role. If the tax

is of the FTT type, it is very eﬃcient in bringing down volatility and raising tax revenue, but if set

too high, the macroeconomy might also be subject to very strong distortion. The FAT is less able to

stabilize the market and also generates less revenue for the state. But in contrast to the FTT, it does

not create large distortions when set too high. We have shown that the optimal decision of making

the FAT ﬂat or progressive is depending on the tax rate. For values below 11% the progressive

version is the best choice, while for rates above 40% the ﬂat tax version is preferable. In between,

the progressive tax leads to better stabilization while the ﬂat tax generates more revenue.

Our model is stylized and simple to implement. Of course, it can also be used for numerous

augmentations: (1) The eﬀects of diﬀerent cross-sectoral channels (e.g. Tobin’s q or stock wealth

eﬀect) can be analyzed. (2) The rules that deﬁne the behavior of the ﬁnancial market agents

(like the time horizon of investors’ strategies) can be changed. (3) Since the occurrence of bubbles

implies large deviations from the fundamental steady state, one might also use a version of the NKM

submodel that is not log-linearized. (4) Moreover, we do not take ﬁnancial streams between the real

and the ﬁnancial sector explicitly into account. A simpliﬁcation that might be relaxed in future

research.

We see this paper as an early stage in a broader research agenda. The agent-based method

oﬀers enormous new possibilities for macroeconomics. Our research agenda is targeted at further

34

exploration of these possibilities. The paper at hand tries to bridge the gap between this newly

emerging ﬁeld

42

and mainstream macro. It uses new methods (like interaction and evolutionary

learning), but also builds on traditional methods and assumption (like market equilibrium or utility

maximization in the NKM part). Future research will focus on working out the agent-based part

further.

References

Ackerman, F. (2002). Still Dead After All These Years: Interpreting the Failure of General Equilib-

rium Theory, Journal of Economic Methodology 9(2): p. 119–139.

Aoki, M. and Yoshikawa, H. (2007). Reconstructing Macroeconomics: A Perspective from Statistical

Physics and Combinatorial Stochastics Processes, Cambridge University Press.

Bask, M. (2009). Monetary Policy, Stock Price Misalignments and Macroeconomic Instability, Han-

ken School of Economics, Working Papers 540.

Becker, M., Friedmann, R. and Kl¨oßner, S. (2007). Intraday Overreaction of Stock Prices, Saarland

University, Working Papers.

Bernanke, B. and Gertler, M. (1999). Monetary Policy and Asset Price Volatility, Economic Review

4: 17–51.

Bernanke, B. S., Gertler, M. and Gilchrist, S. (1999). The Financial Accelerator in a Quantitative

Business Cycle Framework, In: Handbook of Macroeconomics, Volume I, Edited by J. B. Taylor

and M. Woodford, chapter 21, pp. 1341–1393.

Brock, W. A. and Hommes, C. H. (1998). Heterogeneous Beliefs and Routes to Chaos in a Simple

Asset Pricing Model, Journal of Economic Dynamics and Control 22: 1235–1274.

Caginalp, G., Porter, D. and Smith, V. (2001). Financial Bubbles: Excess Cash, Momentum and

Incomplete Information, Journal of Behavioral Finance 2: 80–99.

Campbell, J. Y., Lo, A. W. and MacKinlay, A. C. (1997). The Econometrics of Financial Markets,

Princeton University Press.

Castelnuovo, E. and Nistico, S. (2010). Stock Market Conditions and Monetary Policy in a DSGE

Model, Journal of Economic Dynamics and Control 34: 1700–1731.

42

See for example Gaﬀeo et al. (2008) or Dosi et al. (2008).

35

Colander, D., F¨ ollmer, H., Haas, A., Goldberg, M., Juselius, K., Kirman, A., Lux, T. and Sloth,

B. (2009). The Financial Crisis and the Systemic Failure of the Economics Profession, Critical

Review 21(2-3): p. 249–267.

Colander, D., Howitt, P., Kirman, A., Leijonhufvud, A. and Mehrling, P. (2008). Beyond DSGE

Models: Toward an Empirically Based Macroeconomics, American Economic Review 98: 236–240.

Dawid, H. and Neugart, M. (forthcoming). Agent-Based Models for Economic Policy Design, Eastern

Economic Journal.

De Bondt, W. F. M. and Thaler, R. (1985). Does the Stock Market Overreact?, The Journal of

Finance 40: 793–805.

De Grauwe, P. (2010a). Animal Spirits and Monetary Policy, Economic Theory.

De Grauwe, P. (2010b). The scientiﬁc foundation of dynamic stochastic general equilibrium (dsge)

models, Public Choice 144: 413–443.

De Grauwe, P. and Grimaldi, M. (2006). Exchange Rate Puzzles: A Tale of Switching Attractors,

European Economic Review 50: 1–33.

Deissenberg, C., van der Hoog, S. and Dawid, H. (2008). EURACE: A Massively Parallel Agent-

Based Model of the European Economy, Applied Mathematics and Computation 204: 541–552.

Delli Gatti, D., Gaﬀeo, E. and Gallegati, M. (2010). Complex Agent-Based Macroeconomics: A

Manifesto for a New Paradigm, Journal of Economic Interaction and Coordination.

Demary, M. (2008). Who Does a Currency Transaction Tax Harm More: Short-Term Speculators

or Long-Term Investors?, Jahrb¨ ucher f¨ ur National¨okonomie und Statistik (Journal of Economics

and Statistics) 228: 228–250.

Demary, M. (2010). Transaction Taxes and Traders with Heterogeneous Investment Horizons in an

Agent-Based Financial Market Model, Economics: The Open-Access, Open-Assessment E-Journal

4.

Dosi, G., Fagiolo, G. and Roventini, A. (2008). The Microfoundations of Business Cycles: An

Evolutionary, Multi-Agent Model, Journal of Evolutionary Economics 18: 413–432.

EU (2010). Financial Sector Taxation, European Commission Services: Working Document

SEC(2010)1166 .

EU (2011a). Investing Today for Growth Tomorrow, Press Releases IP/11/799.

36

EU (2011b). Multiannual Financial Framework (MFF): Questions and Answers, Press Releases

MEMO/11/468.

Fair, R. C. (2009). Has Macro Progressed?, Cowles Foundation Discussion Papers 1728.

Franke, R. and Sacht, S. (2010). Some Observations in the High-Frequency Versions of a Standard

New-Keynesian Model, Christian-Albrechts-Universit¨at Kiel, Economic Working Papers 1.

Frankel, J. A. and Froot, K. A. (1987). Using Survey Data to Test Standard Propositions Regarding

Exchange Rate Expectations, The American Economic Review 77: 133–153.

Gaﬀeo, E., Delli Gatti, D., Desiderio, S. and Gallegati, M. (2008). Adaptive Microfoundations for

Emergent Macroeconomics, Eastern Economic Journal 34: 441–463.

Hommes, C. H. (2006). Heterogeneous Agent Models in Economics and Finance, In: Handbook of

Computational Economics, Volume 2, Edited by Leigh Tesfatsion and Kenneth L. Judd., chap-

ter 23, pp. 1009–1186.

Hommes, C., Sonnemans, J., Tuinstra, J. and van de Velden, H. (2005). Coordination of Expectations

in Asset Pricing Experiments, Review of Financial Studies 18: 955–980.

IMF (2010). A Fair and Substantial Contribution by the Financial Sector, Staﬀ of the International

Monetary Fund: Final Report for the G-20 .

Ito, T. (1990). Foreign Exchange Rate Expectations: Micro Survey Data, The American Economic

Review 80: 434–449.

Keen, M. (2011). Rethinking the Taxation of the Financial Sector, CESifo Economic Studies 57.

Kirman, A. (1993). Ants, Rationality and Recruitment, The Quarterly Journal of Economics

108: 137–156.

Kirman, A. (2010). The Economic Crisis is a Crisis for Economic Theory, Paper presented at the

CESifo Economic Studies Conference on “What’s Wrong with Modern Macroeconomics?”.

Kontonikas, A. and Ioannidis, C. (2005). Should Monetary Policy Respond to Asset Price Misalign-

ments?, Economic Modelling 22: 1105–1121.

Kontonikas, A. and Montagnoli, A. (2006). Optimal Monetary Policy and Asset Price Misalignments,

Scottish Journal of Political Economy 53: 636–654.

Krugman, P. (2009). How Did Economists Get it so Wrong?, New York Times Sept 2.

Lam, J.-P. (2010). The Importance of Commitment in the New Keynesian Model, The B.E. Journal

of Macroeconomics 10.

37

LeBaron, B. (2006). Agent-Based Computational Finance, In: Handbook of Computational Eco-

nomics, Volume 2, Edited by Leigh Tesfatsion ans Kenneth L. Judd., chapter 24, pp. 1187–1233.

Lui, Y.-H. and Mole, D. (1998). The Use of Fundamental and Technical Analyses by Foreign

Exchange Dealers: Hong Kong Evidence, Journal of International Money and Finance 17: 535–

545.

Lux, T. (2009). Stochastic Behavioral Asset-Pricing Models and the Stylized Facts, In: Handbook

of Financial Markets: Dynamics and Evolution. Edited by Thorsten Hens and Klaus R. Schenk-

Hopp´e, chapter 3, pp. 161–215.

Lux, T. and Marchesi, M. (2000). Volatility Clustering in Financial Markets: A Micro-Simulation

of Interactive Agents, International Journal of Theoretical and Applied Finance 3: 675–702.

Lux, T. and Westerhoﬀ, F. (2009). Economics Crisis, Nature Physics 5.

Manski, C. and McFadden, D. (1981). Structural Analysis of Discrete Data with Econometric Ap-

plications, MIT Press, Cambridge.

Matheson, T. (2011). Taxing Financial Transactions: Issues and Evidence, IMF Working Paper 54.

Milani, F. (2008). Learning About the Interdependence Between the Macroeconomy and the Stock

Market, University of California-Irvine, Department of Economics Working Papers 070819.

Nam, K., Pyun, C. S. and Avard, S. L. (2001). Asymmetric Reverting Behavior of Short-Horizon

Stock Returns: An Evidence of Stock Market Overreaction, Journal of Banking & Finance

25: 807–824.

Proa˜ no, C. R. (2011). Exchange Rate Determination, Macroeconomic Dynamics and Stability un-

der Heterogeneous Behavioral FX Expectations, Journal of Economic Behavior & Organization

77: 177–188.

Ravenna, F. and Walsh, C. E. (2006). Optimal Monetary Policy with the Cost Channel, Journal of

Monetary Economics 53: 199–216.

Rudebusch, G. D. (2005). Monetary Policy and Asset Price Bubbles, Federal Reserve Bank of San

Francisco Economic Letter 5.

Samanidou, E., Zschischang, E., Stauﬀer, D. and Lux, T. (2006). Microscopic Models for Financial

Markets, Christian-Albrechts-Universit¨at Kiel, Economic Working Papers 15.

Schulmeister, S., Schratzenstaller, M. and Picek, O. (2008). A General Financial Transaction

Tax: Motives, Revenues, Feasibility and Eﬀects,

¨

Osterreichisches Institut f¨ ur Wirtschafsforschung

Working Paper.

38

Sonnemans, J., Hommes, C., Tuinstra, J. and van de Velden, H. (2004). The Instability of a

Heterogeneous Cobweb Economy: A Strategy Experiment on Expectation Formation, Journal of

Economic Behavior and Organization 54: 453–481.

Taylor, J. B. (1993). Discretion versus Policy Rules in Practice, Carnegie-Rochester Conference

Series on Public Policy 39: 195–214.

Taylor, M. P. and Allen, H. (1992). The Use of Technical Analysis in the Foreign Exchange Market,

Journal of International Money and Finance 11: 304–314.

Tobin, J. (1978). A Proposal for International Monetary Reform, Eastern Economic Journal 4: 153–

159.

Westerhoﬀ, F. (2003). Heterogeneous Traders and the Tobin Tax, Journal of Evolutionary Economics

13: 53–70.

Westerhoﬀ, F. (2008). The Use of Agent-Based Financial Market Models to Test the Eﬀectiveness

of Regulatory Policies, Jahrb¨ ucher f¨ ur National¨okonomie und Statistik (Journal of Economics and

Statistics) 228: 195–227.

Westerhoﬀ, F. H. and Dieci, R. (2006). The Eﬀectiveness of Keynes-Tobin Transaction Taxes When

Heterogeneous Agents Can Trade in Diﬀerent Markets: A Behavioral Finance Approach, Journal

of Economic Dynamics and Control 30: 293–322.

39

Online Appendix

Agent-Based Financial Markets and

New Keynesian Macroeconomics

– A Synthesis –

Matthias Lengnick

∗

University of Kiel

University of Mannheim

Hans-Werner Wohltmann

∗∗

University of Kiel

September 13, 2011

∗

E-mail: matthias.lengnick@vwl.uni-mannheim.de

∗∗

E-mail: wohltmann@economics.uni-kiel.de

1 Robustness to Diﬀerent Deﬁnitions of Weights

For integrating the real- and ﬁnancial sector, we have assumed that all daily realizations of s

t

enter

the quarterly value s

q

with equal weights. An alternative would be to assume geometrically decaying

weights. Equation (24) becomes:

s

q

=

64q

t=64(q−1)+1

weight

t

s

t

(1)

with: weight

t

= (1 − ρ) · ρ

64q−t

0 < ρ < 1 (2)

High values for ρ lead to slowly decaying weights, low values result in quickly decaying weights.

Geometric weights of this form add up to one for very long time periods. For a smaller period of

64 days they do not. We thus rescale our weights by multiplying with a constant that keeps the

relative weight between diﬀerent s

t

constant but yields

64q

t=64(q−1)+1

weight

t

= 1 (3)

Table 1: Robustness check of weighting assumption in equation (24)

Weights equal geometrically decaying

ρ = 0.99 ρ = 0.9 ρ = 0.75 ρ = 0.5

min

τ

vol(x) 0.211 0.211 0.211 0.211 0.211

arg min

τ

vol(x) 0.28% 0.28% 0.28% 0.28% 0.28%

min

τ

dis(x) 0.249 0.249 0.250 0.251 0.251

arg min

τ

dis(x) 0.28% 0.28% 0.28% 0.28% 0.28%

min

τ

Revenue 3.2 ·10

−5

3.2 ·10

−5

3.2 ·10

−5

3.2 ·10

−5

3.2 ·10

−5

arg min

τ

Revenue 0.26% 0.26% 0.26% 0.26% 0.26%

Table 1 shows that our results are very robust against the assumption of diﬀerent weights in (24).

The minimal distortion of the output gap changes marginally. All other results stay constant.

1

1.1 Taylor Rule

Variation A of the Taylor rule has been deﬁned as:

i

q

= λ

δ

π

ˆ

E

q

[π

q+1

] + δ

x

ˆ

E

q

[x

q+1

]

+ (1 − λ)i

q−1

+ ǫ

i

q

(4)

Table 2 shows the results of a robustness check for diﬀerent values of λ. Moving away from our

baseline calibration (λ = 1.0) monotonically increases distortion of x in the optimum. This result

obviously follows from the fact that the central bank is less concerned with stabilizing output and

more with smoothing the interest rate. The volatility of x, instead, falls until λ = 0.75 and rises

afterwards. A moderate weight on interest smoothing therefore reduces volatility of output while a

high weight leads to an increase. The reason for this result is the following. The Taylor rule depends

on expectations about the future values of x and π. Since these are formed in a non-rational way,

they are a potential source of volatility. Reducing the weight of these expectations therefore leads

to a decline of vol(x). If λ becomes too small, the inﬂuence of the stabilizing role of monetary policy

declines by so much that vol(x) rises again.

Table 2: Robustness check of the interest smoothing parameter λ

λ = 1.0 λ = 0.9 λ = 0.75 λ = 0.5 λ = 0.3

min

τ

vol(x) 0.211 0.206 0.203 0.212 0.235

arg min

τ

vol(x) 0.28% 0.27% 0.24% 0.22% 0.14%

min

τ

dis(x) 0.249 0.257 0.278 0.355 0.543

arg min

τ

dis(x) 0.28% 0.27% 0.24% 0.11% 0.08%

min

τ

Revenue 3.2 ·10

−5

3.2 ·10

−5

3.2 ·10

−5

3.3 ·10

−5

3.5 ·10

−5

arg min

τ

Revenue 0.26% 0.26% 0.26% 0.25% 0.18%

Concerning the tax rates, we ﬁnd again that an optimal value might lie signiﬁcantly below the

values that have been found to be optimal in our baseline calibration. Figures 1 and 2 show the

impulse responses under our alternative Taylor rules.

2

0 10 20 30

−1

−0.5

0

0.5

O

u

t

p

u

t

G

a

p

0 10 20 30

−1

−0.5

0

0.5

S

t

o

c

k

P

r

i

c

e

0 10 20 30

−0.5

0

0.5

1

I

n

f

l

a

t

i

o

n

0 10 20 30

−1

−0.5

0

0.5

1

R

e

a

l

I

n

t

e

r

e

s

t

R

a

t

e

Figure 1: Cost shock for i

q

= δ

π

ˆ

E

q

[π

q+1

] + δ

x

ˆ

E

q

[x

q+1

] + ǫ

i

q

0 10 20 30

−2

−1

0

1

2

O

u

t

p

u

t

G

a

p

0 10 20 30

−2

−1

0

1

2

S

t

o

c

k

P

r

i

c

e

0 10 20 30

−0.5

0

0.5

1

1.5

I

n

f

l

a

t

i

o

n

0 10 20 30

−1

−0.5

0

0.5

1

R

e

a

l

I

n

t

e

r

e

s

t

R

a

t

e

Figure 2: Cost shock for i

q

= λ

δ

π

ˆ

E

q

[π

q+1

] + δ

x

ˆ

E

q

[x

q+1

]

+ (1 − λ)i

q−1

+ ǫ

i

q

with λ = 0.3

3

No 2011-09

agent-based financial markets and new keynesian macroeconomics -a synthesis-updated versionby Matthias Lengnick and Hans-Werner Wohltmann

**Agent-Based Financial Markets and New Keynesian Macroeconomics – A Synthesis –
**

Matthias Lengnick∗ University of Kiel University of Mannheim September 13, 2011 Hans-Werner Wohltmann∗∗ University of Kiel

Abstract We combine a simple agent-based model of ﬁnancial markets and a New Keynesian macroeconomic model with bounded rationality via two straightforward channels. The result is a macroeconomic model that allows for the endogenous development of business cycles and stock price bubbles. We show that market sentiments exert important inﬂuence on the macroeconomy. They introduce high volatility into impulse-response functions of macroeconomic variables and thus make the eﬀect of a given shock hard to predict. We also analyze the impact of diﬀerent ﬁnancial transaction taxes (FTT, FAT, progressive FAT) and ﬁnd that such taxes can be used to stabilize the economy and raise funds from the ﬁnancial sector as a contribution to the costs produced by the recent crisis. Our results suggest that the FTT leads to higher tax revenues and better stabilization results then the FAT. However, the FTT might also create huge distortion if set too high, a threat which the FAT does not imply.

JEL classiﬁcation: E0, E62, G01, G18 Keywords: Agent-based modeling; stock market; New Keynesian macroeconomics; ﬁnancial transaction tax; ﬁnancial activities tax

∗ ∗∗

E-mail: matthias.lengnick@vwl.uni-mannheim.de, Tel: +49 621 181 1897, Fax: +49 621 181 1774 E-mail: wohltmann@economics.uni-kiel.de

Economists [...] have to do their best to incorporate the realities of ﬁnance into macroeconomics. Paul Krugman (2009)

1 Introduction

The economies of almost every country have recently been hit by a turmoil in the ﬁnancial markets. This so-called ﬁnancial crisis has vividly demonstrated that developments in the ﬁnancial markets can have major impacts on the real economy. Interdependencies between real and ﬁnancial markets should therefore obviously be taken into account when doing macroeconomics. Natural questions to ask after the recent crisis are: To which extent does the formation and bursting of bubbles spill over into real markets? Can ﬁnancial market regulation be used to reduce disturbances of the real economy? How can the ﬁnancial sector be hold to account for the enormous costs created by the recent crisis? For about two decades now, a relatively new modeling approach has been applied to the analysis of ﬁnancial and foreign exchange markets. This approach builds on the method of agent-based computational (ABC) simulation, it drops the assumptions of rational expectations, homogeneous individuals, perfect ex ante coordination and often also market equilibria, in favor of adaptive learning, simple interactions of heterogeneous agents, and emerging complex macroscopic phenomena.1 The approach seems very promising thus far since, on the one hand, it is grounded in the results of survey studies2 and laboratory experiments3 , and on the other hand, the emerging macro-dynamics mimic the properties of real world data (such as martingale property of stock prices, fat tails of return distribution, volatility clustering and dependency in higher moments)4 quite well, a success that traditional ﬁnancial market models, building on equilibrium and rationality, do not provide.5 A huge literature has already developed on this topic that – despite its success – is largely ignored by macroeconomists.

1

2 3 4 5

For an introduction into ABC ﬁnancial market modeling see, e.g., Samanidou et al. (2006), Hommes (2006) or LeBaron (2006). Outstanding examples of such models are Kirman (1993), Brock and Hommes (1998), and Lux and Marchesi (2000). Consult Frankel and Froot (1987), Ito (1990), Taylor and Allen (1992) and Lui and Mole (1998). Consult Caginalp et al. (2001), Sonnemans et al. (2004) and Hommes et al. (2005). A detailed description of these stylized facts can be found in Lux (2009). De Grauwe and Grimaldi (2006), for example, compare the performance of an agent-based model with popular models like that of Obstfeld and Rogoﬀ in explaining the stylized facts of foreign exchange rates. They ﬁnd that the former performs much better.

1

Their aim is to provide insights into the dynamics of the stock price component that is driven by utility-optimizing. 2 . The decision which strategy to employ is reached by using an evolutionary approach: A continuous evaluation of those strategies according to past performance leads to changes in the size of the diﬀerent investor groups. convergence towards the fundamental value sets in. Lux and Westerhoﬀ (2009). In such models. a number of authors are calling for the use of ABC models in macroeconomics. Examples of purely agent-based macro models (with no connection to NKM) are Gaﬀeo et al.6 According to them. each bubble has a constant exogenous probability to burst. Some macroeconomic models already allow for such misalignments. Stock price dynamics in such models are a rational response of an ex ante perfectly coordinated economy in equilibrium. Kontonikas & Ioannidis [KI] (2005) and Kontonikas & Montagnoli [KM] (2006) use forward. it is hard to imagine stock price bubbles or ﬁnancial crisis in such frameworks. augment the model of Bernanke et al. (2010).One strength of the ABC method is that it naturally allows for the endogenous emergence of bubbles. bubbles) on the ﬁnancial markets is often seen as having the most devastating impact on the real economy. the assumptions of equilibrium. In their model. e. Inspired by the spectacular failure of mainstream macroeconomics to provide an explanation of the current crisis and an agenda of how to deal with it. With the mentioned criticism in mind. (2009). and Dawid and Neugart (forthcoming). investors can typically choose from a set of diﬀerent non-rational trading strategies. The emergence of asset price misalignments (i. Colander et al. Bernanke and Gertler (1999). Kirman (2010).and backward-looking New Keynesian macroeconomic (NKM) models with lagged stock wealth eﬀects. where ”burst” simply means that asset prices immediately return to their fundamental value. stock prices can deviate sharply from their underlying fundamental value. Instead they make use of an endogenous dynamic process that binds stock prices to two diﬀerent forces: One of which leads to a return towards the fundamental 6 See. Krugman (2009). for example. In phases that are dominated by technically operating investors.. Colander et al. rational-expecting agents. perfect ex ante coordination. (1999) by imposing an exogenously given path for asset price misalignment.e. (2008) or Deissenberg et al. rational expectations and representative agents are very unrealistic and the reason that macroeconomists have become blind to crisis.g. If market sentiments change and fundamentalists dominate. (2008). Stock price dynamics in these models are not exogenously imposed and the crash of a bubble does not simply occur with a ﬁxed probability. More recently Milani (2008) and Castelnuovo and Nistico (2010) have integrated stock price misalignment into a New Keynesian DSGE model. Delli Gatti et al. (2008).

Such a comprehensive model allows for the simultaneous development of endogenous business cycles and stock price bubbles. To the best of our knowledge. In section 4 we analyze the 7 By market sentiments we mean the state of agents opinions about economic variables that are the result of an evolutionary process and not of rational forecasting. that is so important for ABC ﬁnancial markets. We use our model to answer two questions that are currently on the international policy agenda: (1) Is a Financial Transaction Tax (FTT) or a Financial Activities Tax (FAT) better suited for regulating ﬁnancial markets and generating tax income for the state? (2) Should the rate of a FAT be ﬂat or progressive? The model is developed in section 2. Nonetheless. We demonstrate the working of our model by means of numerical simulation and impulse response analysis in section 3. We also ﬁnd that the negative impact that speculative behavior of ﬁnancial market participants exerts on the macroeconomy. we merge a simple ABC model of ﬁnancial markets with the New Keynesian DSGE model. our approach focuses on simplicity.value. and the other – so-called momentum eﬀect – relates stock prices to their own past development. none of the above models explicitly motivates the dynamics of stock price misalignment by boundedly rational investor behavior and none makes use of an evolutionary selection mechanism of trading strategies that is used in ABC type models. While KI (2005) and KM (2006) are clearly inspired by the agent-based ﬁnancial markets literature with its fundamentalist and chartist trading rules. Expectations in both submodels are formed by an evolutionary selection process. no such attempt has been made so far. our model leads to a number of interesting insights. and do not want to exclude readers who are not familiar with both of these areas. is missing. 3 . Since we combine two separate subdisciplines of economics. it does not allow for an endogenous evaluation of the diﬀerent investment strategies. In this paper. Investors therefore keep employing the same investment rule and do not try to learn from past observations. While the model provides the major advantage that it justiﬁes stock price movements by the behavior of these two types of investors. As in KI (2005) and KM (2006) the aspect of evolutionary learning. can be reduced by introducing a tax on ﬁnancial transactions. We ﬁnd that the transmission of shocks is dependent on the state of market sentiments7 at the time of its occurrence. In a recent paper Bask (2009) uses a New Keynesian dynamic stochastic general equilibrium (DSGE) framework with stock prices that are determined by the demand of two diﬀerent types of investors: chartists and fundamentalists. The expression is taken from De Grauwe (2010a) where it is used as a synonym to animal spirits.

and one the real sector of the economy. Section 6 concludes. 10 On the explicit modeling of high frequency New Keynesian models see Franke and Sacht (2010). a large fraction of ﬁnancial transactions (10%-60% according to market)9 are accounted by algorithmic trading which is typically of an extremely short-term intra-daily nature. our way of integrating the ﬁnancial market and the real economy is very diﬀerent. who ﬁnd that time series of real economic data do not share the power law distribution of ﬁnancial markets which implies that the latter are characterized by higher economic activity. In order to prohibit contradictions stemming from these diﬀerent assumptions of 8 Although this argument seems to be straightforward it is also backed empirically by Aoki and Yoshikawa (2007). We use the ABC chartist-fundamentalist model proposed by Westerhoﬀ (2008) to model the ﬁnancial market. The paper at hand can thus be seen as a complementary approach to a similar research target. 9 Consult Matheson (2011). p. The real sector is described by the NKM framework augmented by a cost eﬀect of stock prices. Our ﬁnancial market. Second. This implies that both can not be modeled on the same time scale. Since we allow for an endogenous development of business cycles and stock price bubbles. An approach which is related to ours can be found in Proa˜o (2011). but impose their dynamics exogenously (section 1). ﬁrst. 19.8 For example.10 The two modeling methodologies employed throughout this paper are building on very diﬀerent assumptions. 4 . The author makes use of an n ACE foreign exchange market in the context of an open economy macro model. It is also an augmentation of those models that integrate a stock market with diﬀerent types of investors into macroeconomics. While the general idea is similar to ours. it operates on a smaller time interval than the real economy. our model is an augmentation of NKM models that already include stock price bubbles. Economic transactions in the former seem to take place much more frequently than in the latter. in contrast to Proa˜o (2011). The ﬁrst problem one has to deal with is that the rules determining the dynamics of ﬁnancial markets are likely to be very diﬀerent from those of the real markets.introduction of diﬀerent kinds of taxes levied on ﬁnancial transactions. one describing the ﬁnancial sector. 2 The Model Our model consists of two parts. Our results are checked for robustnes in section 5. but do not employ endogenous learning (section 1). contains noise and is therefore n not completely deterministic.

Subsection 2. real market agents have neither detailed knowledge about ﬁnancial markets. but take the diﬀerences seriously.13 Since fundamentalist and chartist investment strategies do not account for all possible strategies that exist in real markets. for example. and a is a positive reaction parameter. 2. respectively.11 In this model. Demary (2010) also analyzes the eﬀects of introducing such taxes in a basic Westerhoﬀ-model augmented by diﬀerent time horizons of investors. because it has already been used for policy analysis (especially transaction taxes) so that its behavior in this respect is well known. Conversely.i. ¯ the stock price st also equals the stock price misalignment. also used in Westerhoﬀ and Dieci (2006) who model two ﬁnancial markets and their interaction when introducing transaction taxes. 12 Negative orders denote a supply of stock. 5 . we make use of the standard assumption that the true (log) fundamental value of the stock price sf equals zero. Thus.d. (1) can be interpreted as a market maker scenario. As a result. Subsection 2. a noise term ǫs is added that is i. nor the possibility to participate in high frequency trading. stock price adjustment is given by a price impact function: C F st+1 = st + a WtC Dt + WtF Dt + ǫs t (1) DC and DF stand for the orders generated by chartists and fundamentalists. normally distributed with standard deviation σ s .2 deﬁnes the real one. while 2.12 W C and W F denote the fractions of agents using these strategies. It can be interpreted as the inﬂuence of those other strategies. For the sake of simplicity. we must assume that real and ﬁnancial markets are populated by diﬀerent kinds of agents. who have the resources to participate in high frequency trading.3 brings the two sectors together.ABC and DSGE modeling. Eq. and second.1 deﬁnes the ﬁnancial sector of our economy.1 Financial Market We use the model proposed by Westerhoﬀ (2008) to deﬁne the ﬁnancial sector of our economy for two reasons: First. where prices are adjusted according to observed excess demand. We interpret those of the ﬁnancial market to be institutional investors. we do not simply integrate one into the other. because of its straightforward assumptions and easy implementation. 13 There are also agent-based ﬁnancial models that make use of Walrasian market clearing. See for example Brock and Hommes (1998). t denotes the time index which is interpreted as days. 11 The approach is.

The demand generated by chartist and fundamentalist trading rules is therefore given by:14 C Dt = b (st − st−1 ) + ǫC t F D t = c sf − st + ǫF t t b = ℓ · kC c = ℓ · kF (5) (6) The fractions of agents using the two diﬀerent investment strategies are not ﬁxed over time. 6 . Since (2) and (3) do not reﬂect the great amount of chartist and fundamentalist trading strategies that exist in real world markets. agents continuously evaluate the strategies they use according to past performance. the noise term ǫi is added. Fundamentalists. on the other hand. t It is normally distributed with standard deviation σ i and can be interpreted as the inﬂuence of all other forecasting strategies diﬀerent from (2) and (3). expect that k F · 100 % of the actual perceived mispricing is corrected during the next period: ˜F Et [st+1 − st ] = k F sf − st t (3) sf is the perceived fundamental value that does not necessarily equal its true counterpart sf . The ¯ t diﬀerence between sf and sf is explained in detail in subsection 2. The tilde on the expectations operator indicates that the expectation is not formed rationally. It 14 Westerhoﬀ (2008) directly assumes eq. the more likely it is that agents will employ it.Chartists expect that the direction of the recently observed price trend is going to continue: ˜C Et [st+1 − st ] = k C [st − st−1 ] (2) k C is a positive parameter that denotes the strength of trend extrapolation. The better a strategy performs relative to the other. Instead.3. (5) and (6) and does not explicitly state the diﬀerent types of expectation formations. F } (4) ℓ is a positive reaction parameter. Assuming that the demand ¯ t generated by each type of investors depends positively on the expected price development leads to: i Dt = ℓ Ei [st+1 − st ] + ǫi t t i = {C.

Their proﬁtability ultimately depends on the price realization in t. the higher the fraction of agents using it. In order to calculate nominal proﬁts. st has to be delogarithmized. F } (7) The memory parameter 0 ≤ d ≤ 1 deﬁnes the strength with which agents discount past proﬁts. 0} (8) The more attractive a strategy. 7 . This parameter is often called the rationality parameter in ABC ﬁnancial market models. Hommes (2006) and Westerhoﬀ (2008). sf . Manski and McFadden (1981) for a detailed explanation of discrete choice models. The fraction of agents t t that employ strategy i is given by the well known discrete choice or Gibbs probabilities:16 Wti = exp{eAi } t C } + exp{eAF } + exp{eA0 } exp{eAt t t i = {C. 17 Consult Westerhoﬀ and Dieci (2006).g. The extreme cases d = 0 and d = 1 relate to scenarios where agents have zero and inﬁnite memory. Note that the probability of choosing one of the three strategies is bounded between zero and one. The only diﬀerence between our ﬁnancial market submodel and that of Westerhoﬀ (2008) is that we distinguish between the true fundamental value sf and the trader’s perception of it. the greater (lesser) the fraction of agents that will employ the strategy with the highest attractiveness. The positive parameter e measures the intensity of choice. Agents may also withdraw from trading (strategy “0”). See. The higher (lower) e..17 The described mechanism can be interpreted as an evolutionary survival of the most proﬁtable forecasting strategy. Both ¯ t models are equivalent if sf = sf . The attractiveness of this strategy A0 is normalized to zero A0 = 0 .is assumed that the attractiveness of a particular strategy depends on its most recent performance i (exp{st } − exp{st−1 }) Dt−2 as well as its past attractiveness Ai :15 t−1 i Ai = (exp{st } − exp{st−1 }) Dt−2 + dAi t t−1 i = {C. Note the timing of the model: Orders submitted in t − 2 are executed in t − 1. F. e. ¯ t 15 16 Recall that st is the logarithm of the stock price.

.e. x the (log) output gap (i.. we use equal expectations formation for the central bank and the market. agents could use assets they hold as collateral when borrowing money. its deviation from steady state). ǫπ are stochastic elements with q q q zero mean. The central bank reacts to expected deviations of inﬂation and output from its target. In this context. The dynamic path of the stock price s is determined by the model developed in the previous subsection. Equation (10) is referred to as the dynamic IS-curve that describes the demand side of the economy. . In section 5. ˜ ˜ iq = δπ Eq [πq+1 ] + δx Eq [xq+1 ] + ǫi q ˜ xq = χEq [xq+1 ] + (1 − χ)xq−1 − ˜ πq = β ψ Eq [πq+1 ] + (1 − ψ)πq−1 1 ˜ iq − Eq [πq+1 ] σ + ǫx q (9) (10) (11) + γxq − κsq + ǫπ q The notation of the variables is as follows: i is the deviation of the nominal interest rate from its target. Asset prices inﬂuence the economy through a balance sheet channel that works as follows: The willingness of banks to grant credits typically depends on the borrowers’ ﬁnancial position. For example. The more collateral a debtor has to oﬀer.2 we will generalize the model by assuming that the central bank’s expectations are diﬀerent from those of the market. We keep the common ¯ interpretation of the time index in New Keynesian models and assume that it denotes quarters. and s the deviation of the (log) nominal stock price from its true fundamental value sf . where the tilde indicates that they are formed non-rationally. ˜ Eq [·] is the expectations operator conditional on knowledge available in q. π the deviation of the inﬂation rate from its target. The variables ǫi . For now. Equation (9) is a standard monetary policy interest rule. The subscript q = 1. It can be derived under the assumptions of nominal price rigidity and monopolistic competition. New Keynesian models are widely used in macroeconomics because they typically allow for a good ﬁt of real world data. Equation (11) is a New Keynesian Phillips curve that represents the supply side. Q denotes the time index. “advantageous” may 8 .2 Real Markets The partial model describing the real sector is given by a hybrid NKM model. and they are derived from individual optimization. It results from the Euler equation (which is the result of intertemporal utility maximization) and market clearing in the goods market.. ǫx .2. the more advantageous his credit contract will be.

Kontonikas and Ioannidis (2005). and allow them access to cheaper credits. it is commonly assumed that the household’s only possibility of transferring wealth into future periods is by demanding bonds. Consequently. Note also that our deﬁnition of the stock price gap is very diﬀerent from that of Milani (2008) or Castelnuovo and Nistico (2010). +κsq ). ABC ﬁnancial market models could also be employed for the analysis of foreign exchange rates.18 This verbal kind of micro foundation is suﬃcient for our purposes. To avoid confusion. the cost channel is typically introduced by adding interest costs (+α · iq ) into the Pillips-Curve. The ﬁrst argument can be used to relate asset prices positively to aggregate demand. who deﬁne it as the diﬀerence between the stock price under fully ﬂexible and somewhat rigid market conditions. Analogously to this channel. Since a rise (fall) of foreign exchange rates would also raise (lower) production costs – via more expensive (cheaper) intermediate inputs – they would be included with the opposite sign (i. Kontonikas and Montagnoli (2006). are the result of utility optimal paths under rational expectations. Note that we deﬁned sq as the nominal stock price gap. If one agent wins from a 18 Formally. Households therefore do not hold or trade stock. 9 . asset prices are inversely related to ﬁrms marginal (real) costs of production. The ﬁnancial sector can not generate proﬁts on the aggregate level by selling and reselling stock. We further assume that ﬁrms hold an initial amount of stock but do not participate in stock trading. The so-called cost channel of monetary transmission is commonly introduced into New Keynesian models by adding the nominal interest rate into the Phillips-curve (see for example Ravenna and Walsh (2006) or Lam (2010)). they are only aﬀected by the ﬁnancial sector via the balance sheet channel. and not via speculative gains. we also decided to insert the nominal (and not the real) stock price gap into (11). We keep this assumption in order to allow for analysis of the isolated impact of the speculation of ﬁnancial market participants on stock prices. This argument allows the addition of the term −κsq to equation (11). Since most ﬁrms’ production is largely ﬁnanced through credits. (10). Higher prices of assets owned by ﬁrms increase their creditworthiness. The reader is referred to Bernanke and Gertler (1999) who discuss a balance sheet channel (and its microfoundation) in more detail. we want to point out again that we are modeling stock prices with the ABC submodel and not foreign exchange rates. as for example done in Bernanke and Gertler (1999). We stress the second argument in this paper. the term −κ · sq has to be added instead. If interest costs are negatively depending on solvency and thus stock prices. To derive eq.mean that either credits of larger size are oﬀered or that credits of the same size could be obtained cheaper (lower interest payments). or Bask (2009). of course.e. Both.

ﬁnancial streams between the real and ﬁnancial sector do not exist. Because their relative size is small when calculated for a daily basis. πq−1 ). and because the Westerhoﬀ-model does not explicitly take ﬁnancial wealth into account. As a result of the above arguments and assumptions. Expectations in the real sector are also formed in a non-rational way. q q q The endogenous real sector variables iq . ǫπ ). The only possibility for the aggregate stock market to earn proﬁts is by dividend payments from the real sector. the current value of sq and the realizations of the noise terms (ǫi . x and π depend on the future expectations Eq [xq+1 ]. Following De Grauwe opt (2010a) and De Grauwe (2010b) we assume that a certain fraction ωq of agents is optimistic about 10 .beneﬁcial transaction. their own history (iq−1 . We take a closer look on the stability conditions of the system in the next subsection after the model has been stated completely. ˜ Eq [πq+1 ]). xq−1 . we do not model the stream of dividend payments from ﬁrms to ﬁnancial investors. xq and πq can be calculated as follows: iq 0 iq−1 0 ǫi q ˜ xq = A−1 B Eq [xq+1 ] ˜ Eq [πq+1 ] πq + A−1 C xq−1 πq−1 + A−1 0 sq + A−1 ǫx q −κ ǫπ q (13) Of course the parameters must be selected in a way that the system is stable. others must lose. ǫx . The model can be rearranged as follows: 1 1 σ 0 1 0 0 0 xq = 0 −γ 1 πq 0 A iq 0 δx χ 0 B ˜ Eq [xq+1 ] ˜ Eq [πq+1 ] βψ 1 σ δπ 0 + 0 1−χ 0 0 0 β(1 − ψ) C 0 0 0 xq−1 πq−1 iq−1 0 ǫi q + 0 sq + ǫx q ǫπ −κ q (12) ˜ The dynamics of the forward-looking variables i.

11 . the weights that determine the fractions of agents are given by: opt ωq = exp{φAopt } q exp{φAopt } + exp{φApes } q q pes exp{φAq } exp{φAopt } + exp{φApes } q q (19) (20) and pes ωq = 19 In all numerical simulations we set µ = 0. we deﬁne the attractiveness of the diﬀerent strategies as: ˜ opt Aopt = − xq−1 − Eq−2 [xq−1 ] q ˜ pes Apes = − xq−1 − Eq−2 [xq−1 ] q 2 2 + ζAopt q−1 + ζApes q−1 (17) (18) The attractiveness of forecasting strategies are therefore determined by past mean squared forecast errors (MSFEs) weighted with decaying weights. Applying discrete choice theory. Both groups form expectations according to: ˜ opt Optimists expectation: Eq [xq+1 ] = gt ˜ pes Pessimists expectation: Eq [xq+1 ] = −gt ˜ opt = −Eq [xq+1 ] (14) (15) The spread between the two expectations (2gt ) is assumed to vary over time according to: 2gt = µ + ν · Std [xt ] (16) The parameters satisfy µ.19 The economic rationale behind this implementation is that the agents beliefs diverge more when uncertainty surrounding the output gap is high.pes the future development of output while another ωq is pessimistic. In line with the method provided in the previous subsection. The mentioned time windows is set to 20 periods. In the special case of ν = 0 the divergence of beliefs is constant over time.5 and ν = 2. ν ≥ 0 and Std [xt ] denotes the unconditional standard deviation of the output gap computed over a ﬁxed window of past observations.

20 Consult Ackerman (2002). The diﬀerence to conventional rational expectations (RE) is that no single agent is required to expect future dynamics rationally. It is implausible that real world human beings are capable of solving such highly complex problems. agents choose from a set of forecasting rules that are so simple that real world human beings would be able to employ them. are not unrational. Where the fractions of targeters and extrapolators (ω tar and ω ext ) are again determined by the same evolutionary approach used for expectations about the output gap. Expectations of this group are given by: ˜ ext Eq [πq+1 ] = πq−1 (23) ˜ The markets expectation Eq [πq+1 ] is again determined as the weighted average of these two groups. they extrapolate past values into the future. Using such simple rules is not unrational. 12 . i. One type of agents (the targeters) believes in the inﬂation target that the central bank has announced. hence their expectations are given by: ˜ tar Eq [πq+1 ] = π ⋆ (22) Another group (the extrapolators) expect that the future inﬂation rate is given by the most recently observed one. opt ˜ opt pes ˜ pes ˜ Eq [xq+1 ] = ωq Eq [xq+1 ] + ωq Eq [xq+1 ] (21) Expectations about the inﬂation rate in De Grauwe (2010a) and De Grauwe (2010b) are formed in a similar way.The market’s expectation of the output gap is given by the weighted average of the two diﬀerent forecasting strategies. Gaﬀeo et al. it can be understood as the best way to deal with an overwhelmingly complex world. Eq [xt+1 ] and Eq [πt+1 ]. In our model.20 It would require every agent to know how everybody else would react in every possible situation and to calculate the resulting mean time paths in advance. An evolutionary mechanism is used to permanently evaluate these strategies and sort out the poorly performing in favor of the better ones. that forming conventional RE would indeed be impossibly complicate. Fair (2009) and Kirman (2010).e. It has been pointed out by a number of authors. (2008). ˜ ˜ Both expectations.

64q. the two parts of the model run on diﬀerent time scales.3 Bringing the Two Sectors Together As already mentioned. 2. the ﬁnancial sector performs 64 increments of the time index t within one increment of the real market’s time index q (ﬁgure 1). instead of requiring rationality from the individuals (as conventional rational expectations do).Hence.. We assume that one quarter consists of 64 trading days. Quarter q is deﬁned to contain the days 64(q − 1) + 1 .. The real markets operate quarterly while the ﬁnancial market operates daily. . . it is the result of an evolutionary dynamic market process. Therefore. .

One could instead also introduce a discounting factor into (24) to raise the relative inﬂuence of the more recent days.1. Figure 1: Time scale as indexed by days (t) and quarters (q) We assume that the relevant value of the quarterly stock price sq that aﬀects the real sector via eq. q · 64) with the agent-based model deﬁned in section 2. The fundamental s t 21 Eq. and insert the mean of the resulting st ’s into eq. . (13) is the average of the daily realizations of st of the corresponding quarter q.. Now that we have set up the real and ﬁnancial markets we can deﬁne the diﬀerence between the true fundamental stock price (¯f ) and the fundamentalist’s perception of it (sf ). (13) in order to get the impact on real sector variables. . 13 .. we calculate the recursive dynamics of the ﬁnancial market for one quarter q (in days: t = (q − 1) · 64 + 1 . Thus sq is given by:21 1 sq = 64 64q st t=64(q−1)+1 (24) Using the deﬁnitions above. We show in the online appendix that our results depend only marginally on the decision to use such a discounting factor. (24) assumes that the inﬂuence of daily stock prices on the real economy is equal for each day in the quarter.

.

.

.

.

.

.

.

Therefore.e. (25)). i. given that agents behave rationally. h≥0 (26) The ﬂoor-function rounds a real number down to the next integer. i. it has been argued that in reality it is very diﬃcult (if not impossible) to identify the true fundamental value of any stock. Figure 2: Channels between real and ﬁnancial markets value of any given stock is commonly understood to be the sum of all discounted future dividend payments dt+k . 24 For example Rudebusch (2005) or Bernanke and Gertler (1999) raise doubts of this kind. since the latter is not modeled endogenously.e. Most models do 22 23 Consult Campbell et al. Eq. De Bondt and Thaler (1985) were among the ﬁrst to describe this phenomenon. but instead simply take the current development of the real economy as a proxy for it. if t output is high (low) the fundamental stock price is perceived to lie above (below) its true counterpart. We decided to model the perception of the fundamental value in a diﬀerent way for two reasons: First. In the most simple case it could be given by something similar to:22 ∞ sf = t k=1 ρk Et [dt+k ] (25) Dividends are typically closely related to real economic conditions (xq in our model). stock prices show stronger reactions to new information than they should.23 Second. (26) states that the fundamentalists’ perception sf is biased in the direction of the most recent real economic activity. Note that ABC models of ﬁnancial markets can typically not relate the fundamental value to the recent economic development. it has been empirically found that stock markets overreact to new information. s f = h · xq t q = ﬂoor t−1 64 .24 Given these problems. 14 . (1997) chapter 7 for the derivation of this equation and more general versions. it seems reasonable to assume that agents do not know the true value of sf or calculate ¯ it in a rational way (as in eq. sf t would depend on the expectation of x for all future days.

Figure 3(a) shows a numerical approximation of the stability region in h-κ-space. If stock prices are high. It is known that the policy parameters δπ and δx are crucial for the stability of the NKM model. 25 26 Again. (11) is set equal to κ = 0. Channel II (the misperception of sf channel) allows for inﬂuence ¯ in the opposite direction.5 h 1 1.4 0. If both of these cross-sectoral parameters are set equal to zero (κ = 0 & h = 0). Channel I (the cost channel) allows the ﬁnancial market to inﬂuence the real sector and disappears if κ in eq. Marginal costs and thus inﬂation fall. and so on. both sectors (i. The parameterization used for this numerical investigation is discussed in detail below.6 δπ 0. To exclude explosive paths.e.2 0. we generate a numerical approximation of the stability region in δx -δπ -space (ﬁgure 3(b)).5 2 Stable Region 0. As a result. To check whether our behavioral model possesses a similar property. (25) is set equal to h = 0.1 3 Unstable Region 0. The two cross-sectoral channels feed on each other.6 0. a suﬃcient condition for stability is δπ > 1 and δx ≥ 0. Channel I exerts a positive inﬂuence on output: Solvency of ﬁrms rises which lowers their credit costs.5 0. Westerhoﬀ (2008) is a good example to look at since both of these approaches are discussed there.4 0. both submodels) operate independently of each other. The system is stable for δπ > 1 and δx ≥ 0 and thus features the typical stability properties of NKM models. and disappears if h in eq.5 1 κ Stable Region 0 0 0. κ has to be lower the higher h and vice versa. they set both equal to zero or assume them to follow a random ¯ t walk. output rises.8 1 δx (a) in h-κ-space (b) in δx -δπ -space Figure 3: Stability Analysis26 not distinguish between sf and sf .8 2.25 Figure 2 illustrates the two channels that exist between the real and the ﬁnancial market. which in turn exerts a positive inﬂuence on stock prices through Channel II.5 Unstable Region 0 0 0. 15 .2 1. Under standard speciﬁcation.

when comparing these two values. The values of σ. this implies a strong reaction of s to x and thus a high value of h.15 Interaction κ = 0. Nam et al.5 in order to maintain the endogenous business cycles of the De Grauwe (2010a) model. The self-fulﬁlling character of expectations (explained in detail below) will ultimately break down. For the policy parameters δx and δπ we use the values that have originally been suggested by Taylor (1993). (2001).3 Numerical Simulations The analysis of our model is performed by means of numerical simulation.28 Therefore we set h to be ten times larger than κ.17166 β = 0.01 Real sector σ=1 γ = 0. Bringing it to a quarterly basis we obtain d64 = 0.05 σ F = 0. The calibration is given in Table 1.5 ζ = 0. 27 16 .04 K F = 0.975 e = 300 σ s = 0.5 φ = 10 χ = 0.8. If stock prices over react to new information (De Bondt and Thaler (1985).5 δπ = 1.5.1 h=1 The memory parameter for the ﬁnancial sector d = 0.5 (taken from De Grauwe (2010a)). The hybridity parameters χ and ψ are set to 0. γ and β are so-called ”deep parameters” (or functions of such) and common in New Keynesian models.8 σǫ = 0. Table 1: Baseline Calibration of the Model Financial sector a=1 K C = 0. These parameters have to be set larger than 0.27 In order to set the cross-sectoral parameters.2.8 ψ = 0. 28 We have assumed that the value of x is taken as information for the development of the real sector.99 δx = 0. Becker et al. Therefore our calibration presumes that past information are less taken into account by ﬁnancial market agents than by real sector agents. The parameter values for the ﬁnancial sector are exactly the same as in Westerhoﬀ (2008). However. the two terms containing Eq [xq+1 ] cancel out if χ = 0. it has to be taken into account that d refers to the discounting of information that is only one day old. (2007)).01 σ C = 0.04 ℓ=1 d = 0. Whether agents are optimistic or pessimistic does not play any role for the determination of xq any more. we assume that the real sector is much less inﬂuenced by the ﬁnancial sector than the other way round. The intensity of choice parameters (e and φ) seam most problematic ˜ One can easily see that if (9) is inserted into (10).975 (which is taken from Westerhoﬀ (2008)) is much higher than that for the real sector ζ = 0.

Vice versa for phases dominated by pessimists. cyclical regime switching. gray fundamentalist trading (W F ). Closely connected to the up and down phases of output is the fraction of optimists (ω opt ). Black denotes chartist trading (W C ). the mechanism is not strong enough to generate a self-fulﬁlling. around q = 35). st . ω tar . a series of pseudo random numbers has to be drawn. the situation has a tendency to reproduce itself. A similar pattern can be observed for expectation formation of the inﬂation rate. This result obviously follows from the speciﬁcation of our learning mechanism. Most of the time. expectations about the output gap have a self-fulﬁlling nature and generate business cycles. To eliminate the inﬂuence of arbitrary initial conditions each simulation is performed with a ”burn-in” phase of 20 quarters. Figure 4 shows ˜ the resulting dynamics for xq . As a result. However.1 Dynamics of one Simulation To demonstrate the working of our model. Hence. Hence each realization of simulated data is a unique result of the underlying random seed. again. ω opt . The latter represents the fraction of agents. Hence the model generates an endogenous business cycle. we perform one “representative” run. the fraction of targeters (ω tar ) decreases and inﬂation can become very low. iq − Eq [πq+1 ] . To generate the dynamics.since their impact on the results depends on the latent attractiveness values. Since the resulting high output. employing the three trading strategies. The output gap is characterized by cyclical ups and downs. The simulated time period consists of 100 quarters (or 6400 days). πq . quarters cover an interval of 64 data points. The horizontal time axes are quarterly scaled. the self-fulﬁlling tendency is not as strong as for the output gap. If inﬂation largely deviates from the target (e. This results in a larger fraction of optimists and thus in a market expectation above the steady state of zero ˜ Eq [xq+1 ] > 0 . and white no trading (W 0 ). favors optimist forecasting. 17 . The stock market is also characterized by the emergence of diﬀerent regimes. In contrast to the expectations of the output gap. the forecasting rule of optimists (14) performs much better than the pessimists rule (15). In times of high output. In the diagrams containing daily data. and a variable called market sentiments. the targeter’s expectation rule performs poorly relative to that of the extrapolators. 3.g. We will therefore check the robustness of our obtained result to diﬀerent parameterization of e and φ in section 5. iq .

5 0 -0.2 0 20 40 60 80 100 40 60 80 100 1 0.5 0 -0.5 0 -0.5 -1 0.5 20 40 60 80 100 Stock Price (daily) 1 0.6 0.8 0.4 0.8 0.8 ωopt ωtar 0.6 0.4 0.5 0 -0.2 0 20 40 60 80 100 0.5 0 -0.5 Output Gap Inflation 20 1 0.5 10 20 30 40 50 60 70 sf 80 90 100 Market Sentiments (daily) 1 0.2 0 10 20 30 40 50 60 70 80 90 100 WC WF W0 Figure 4: Model Output for a Time Period of Q = 40 18 .6 0.4 0.0.5 20 40 60 80 100 Real Ineterst Rate 20 40 60 80 100 1 Interest Rate 0.

ǫx and ǫπ . In order to allow for 5 impulse response analysis in a way similar to that typically used in DSGE models. Calculate the diﬀerences between the trajectories of step 1 and 2 which gives the isolated impact of the cost shock. The result is an endogenously occurring business cycle and endogenous stock price bubbles. fundamentalists. such questions are typically analyzed via impulse response functions that try to isolate the eﬀects of an exogenous realization of the stochastic terms ǫi . Each forecasting strategy is able to dominate the market from time to time. transitory cost shock without persistence of size ǫπ.+ = 1. the stock price follows its perceived fundamental value (which is given by the dashed line) closely. for example. The opposite case of chartists driving st down below the fundamental value can be found for example around q = 45.. In q = 62 market sentiments turn around. 95) and low during recessions (q = 10. We focus on the impact of an q q q unanticipated. . 19 . although the underlying value falls. Repeat steps 1-3 10. 4. 20 and q = 59. Deviations from sf during such phases are transitory and small in size. The model generates endogenous waves of optimism and pessimism.. Generate the same dynamics with the same random seed (i.. i.when a certain amount of fundamentalists is present in the market. The course of the stock price is thus largely inﬂuenced by the underlying real economy: st is high during the booms (q = 30. who judge st as extremely over-valued. Generate the model dynamics for one particular random seed.e. become dominating and drive the price down again.e. Diﬀerences are thus not a result of diﬀerent random numbers. It continues to follow a slight upward trend.... 40 and q = 80. Note that the noise terms are identical in both runs. 5 5 3. 70). i. 3.. chartists form the dominating majority and the stock price moves away from sf . but solely due to the imposed shock. we perform the following experiment: 1.. the bubble bursts.+ = 1.2 Impulse Response Analysis In this subsection we analyze the eﬀects of an exogenous shock to the real sector. identical realizations of the pseudo random numbers). stock prices become disconnected from the underlying fundamentals... . If the market sentiments change in favor of the chartists.e. but the evolutionary learning assures that none dominates forever. In DSGE models. a bubble builds up. Around q = 60.. inﬂation-targeting and inﬂation-extrapolation as well as chartism and fundamentalism. 2.000 times. . .. but with ǫπ increased by ǫπ.

5 10 20 30 Real Interest Rate 10 20 30 0 −0. On average.Output Gap 0 Stock Price 10 20 30 0 −0. Figure 5 shows the resulting responses to an exogenous shock of ǫπ.5 0 0 −1 0 0. A huge contractionary shock can break this circle and thus turn the boom into a recession. The top row in ﬁgure 6 shows the eﬀect on inﬂation and stock prices. we generate the impulse response functions of π and s that result if either optimists or pessimists are dominating the market during the shock period.75 as dominated by optimists and a situation in which ωq ≥ 0. such an ampliﬁcation mechanism can 20 . Dashed lines are 95 % quantiles. If pessimists dominate. while the dashed lines represent 95% quantiles. The solid lines illustrate mean responses. The only exception is (by construction of the shock) the reaction of inﬂation on impact. We opt pes deﬁne a situation in which ωq ≥ 0.+ = 1 for our baseline cali5 bration.5 −0. Inﬂation and the real interest rate rise. the amplitude is smaller for both time series. To analyze the source of this high volatility.5 −1 0 1 Inflation 0. the economy is caught in a self-reproducing circle of high output and optimistic expectations. the economy shows the typical stagﬂationary response to the cost shock. The other time series exhibit similarly volatile impulse responses.5 0 10 20 30 Figure 5: Mean response to a exogenous cost shock of size one. The reaction of stock prices is stronger when optimists dominate. illustrate that the reaction of xq can be located anywhere between (a) a strong negative reaction on impact that is accelerated during the subsequent two periods and followed by a hump-shaped path back towards trend and (b) no reaction on impact followed by a slightly positive path in the medium and long run. If the number of optimists is high. while output and the stock price fall. All impulse responses show high volatility. If pessimists form the majority.75 as dominated by pessimists. The quantiles for the output gap. The economic logic underlying this phenomenon is the following. for example.

the higher persistence in this case is also a result of self-fulﬁlling expectations. Since they expect the current state of inﬂation to persist. The same contractionary shock therefore has a higher mean impact during a boom than during a recession.2 −0... they create persistence in the system. If targeters dominate (i.. the inﬂation target of the central bank is credible). Persistence.Shock period dominated by . makes their own forecasting strategy more attractive and the number of extrapolators might increase further. If the latter are dominating.. Whether fundamentalists or chartists dominate the stock market does not play a major role in the transmission of cost shocks... Comparing impulse responses for those cases when targeters dominate with those dominated by extrapolators.1 −0. the impact of the shock is stronger on impact and more persistent for both time series. 0 −0.3 −0.2 0 Stock Prices (daily) 10 20 Optimists Pessimists 30 0. Stock Prices (daily) 1 Inflation 0 −0. The economic rationale is that extrapolators do not generate a mean-reversion.4 0 Stock Prices (daily) 10 20 Targeters Extrapolators 30 0.e.2 −0. 21 .4 0 10 20 Fundamentalist Chartists 30 0. 29 Dominance has been deﬁned analogously to the above case.5 0 0 1 Inflation 10 20 30 Shock period dominated by . in turn.5 0 0 10 20 30 Figure 6: Mean response of output and stock price with initial conditions dominated by diﬀerent groups of agents not emerge and the fall in GDP is much smaller. the persistence of the shock is much lower.5 0 0 1 Inflation 10 20 30 Shock period dominated by .29 yields a similar picture (second row in ﬁgure 6). Therefore.1 −0. 0 −0. Strong believes in a stable system obviously lead to a dampening of shocks.

purchase of assets. several countries directly stumbled from the ﬁnancial. Since high frequency. we can use it to analyze a prevailing question currently debated among policy makers.The high volatility that has been found in the impulse responses of ﬁgure 5 is therefore partly a result of the history dependence. while having no signiﬁcant inﬂuence on long term trading. ﬁscal stimuli. the uncertainty about the impact of the shock increases. because persistence has been a matter of concern in NKM modeling. Since the shock can have a diﬀerent impact on average depending on the initial beliefs of agents. while in standard NKM models persistence is introduced exogenously by assuming a hybrid form. Second. it produced a huge decline in GDP and rise in unemployment. A nice executive summary of arguments in favor and against a FTT can be found in Schulmeister et al. ﬁrst. This point is of interest. In its baseline notation (for χ = 1 and ψ = 1 in (10)-(11)) those models do not produce persistent responses to non-persistent shocks. 4 Taxing Financial Transactions The recent ﬁnancial crisis has created enormous costs in all industrialized economies.31 In approaching this question. As a response to those devastating externalities of the ﬁnancial sector.into the ﬁscal crisis. Long run oriented trading that is based on underlying fundamental values. it would be beneﬁcial to curtail it by introducing a FTT. The traditional way of achieving both would be to levy a ﬁnancial transaction tax (FTT) in the spirit of Tobin (1978). speculative trading is a socially wasteful business. whether it should provide a ﬁnancial contribution to the recently generated costs. second. 22 . First. whether new regulatory policies are needed to stabilize this sector for the future and. (2008). On average. Now that we have gained some understanding of the dynamics of the model. the advanced G20 countries suﬀered a rise in government debt by 40%. De Grauwe argues that the evolutionary learning algorithm produces endogenous persistence. is not aﬀected. the G20 leaders have recently asked the IMF to prepare a report 30 31 Consult IMF (2010).30 Because of such rising debt. direct support and many more. it should be asked. Such a tax would make short term trading less attractive. De Grauwe (2010a) has also analyzed the origin of persistence in his behavioral NKM and ﬁnds that responses maintain persistent even if the hybrid character is turned oﬀ. the ﬁnancial positions of states have been very negatively aﬀected because of several necessary stabilization policies like capital injections.

While policy makers are currently intensively debating the introduction of such taxes. We also report the average tax revenue per agent and day that is given as: 1 T T Tax Revenue = WtF · TaxF + WtC · TaxC t t t=1 (27) Where TaxF is the tax payed by fundamentalists and TaxC the tax payed by chartists. 35 Both measures closely follow Westerhoﬀ (2008).33 one striking aspect of the debate is ”that it is almost entirely unguided by the public ﬁnance literature on the topic – because there is hardly any” 34 . In this section. Another slightly diﬀerent type of tax – called ﬁnancial activities tax (FAT) – might be better suited than the FTT. rate of change) of a time series. we report the average fractions of fundamentalists 1 T T F t=1 Wt and chartists 1 T T C t=1 Wt resulting from a certain tax. We do not use the variance 32 33 The mentioned report is IMF (2010). 19). π} (29) The measure vol(·) denotes the volatility (i. EU (2011a) or EU (2011b). See for example EU (2010).e. Additionally we deﬁne the following two measures:35 1 T −1 T vol(s) = |st−1 − st | t=2 dis(s) = 1 T T |st | t=1 (28) And for quarterly time series: 1 Q−1 Q vol(z) = |zq−1 − zq | q=2 dis(z) = 1 Q Q |zq | q=1 z = {x. the IMF argues that taxing ﬁnancial transaction is generally a feasible policy instrument for achieving this goal and that ”the FTT should not be dismissed on grounds of administrative practicality” (p. diﬀerence to fundamental steady state).e.32 Summarizing the results of this report. 34 Quote taken from Keen (2011). the European Commission and the European Parliament are currently examining weather a ﬁnancial tax should be introduced. we use our model to analyze the advantages and disadvantages of both kinds of taxes with regard to their ability to stabilize markets and to raise ﬁscal income. To evaluate their eﬀect on the variables of interest. Accordingly. However it also argues that the traditional FTT might not be the best instrument to ”ﬁnance a resolution mechanism” and ”focus on core sources of ﬁnancial instability”. 23 . dis(·) measures its distortion (i. Besides the IMF and the G20.on how the ”ﬁnancial sector could make a fair and substantial contribution” in bearing parts of the induced burden.

To introduce it into our model. however. Since tax payments directly reduce the proﬁtability of an investment. These studies. we assume that the tax has to be paid relative to the nominal value traded. buys and sells are equally taxed).36 These studies. very recently proposed FAT. (2) We contrast the classical FTT with the innovative. To avoid confusion we do not use the variance measure. Orders generated in Dt−2 imply nominal transactions of Dt−2 · exp{st−1 } in t − 1 and Dt−2 · exp{st } in t. When calculating the variance. The tax rate τ is applied to the absolute nominal value of both transactions (i. eq. 24 . The introduction of a FTT has already often been analyzed in the ABC ﬁnance literature. Westerhoﬀ (2008) or Demary (2008). focus on short-term volatility and neglect long-term mispricing. since it can lead to the built up and bursting of bubbles and therefore might have the most important impact on the real economy. one would not measure the volatility but rather the mean squared distortion. We run the model for 500 quarters (32. 4.1 Financial Transaction Tax The basic characteristics of the FTT is that it is small in size but levied on a brought basis: the total value of transaction.measure because it interprets volatility via the average squared distance from the mean. (7) changes to:37 i i Ai = (exp{st } − exp{st−1 }) Dt−2 − τ (exp{st } + exp{st−1 }) Dt−2 + dAi t t−1 (30) i i The transaction tax is represented by τ and Dt−2 is the absolute value of Dt−2 . however. (3) We also answer a question that has become very prevailing during the recent ﬁscal crisis: how should a tax be designed in order to yield maximal tax revenues? There has also been a number of empirical studies that investigate the impact of FTTs. Since complete investment consists of two transactions. Our time series show long-lasting deviations from the mean (which we interpret as bubbles or distortion). Examples are Westerhoﬀ (2003). Consult also Westerhoﬀ (2008) for the introduction of an FTT into his model. We take the latter into account.000 days) with diﬀerent values for τ as well as 1000 diﬀerent realizations of the 36 37 See IMF (2010) p.e. the tax also has to be paid twice. 20 for a summary of those studies. Our study adds several aspects to this literature: (1) We do not restrict our analysis to the stabilization of ﬁnancial markets but include real markets as well. limit their attention to the reduction of volatility and distortion of stock prices.

24 0.5 Tax Rate 1 0.5 Tax Rate 1 0.2 0 0.3 0 0.14 0. Increasing the FTT leads to monotonically decreasing volatility of stock prices and inﬂation.015 0.22 0.27 0. The measures vol(x). Tax revenue follows a typical Laﬀer curve: Increasing the tax rate up to τ ≈ 0.135 0 0.5 Tax Rate 1 0 0. dis(x) and dis(π) become quickly very large.5 vol(stock price) 0. the gained tax revenue as well as volatility and distortion of s. 25 .5 Tax Rate 1 Figure 7: Impact of Financial Transaction Tax (FTT) pseudo random number generator for each τ . All four follow a u-shaped pattern with minimum near τ = 0.fundamentalists (solid) chartists (dashed) 0.4 0.136 0. With respect to vol(s) and vol(π). At the same time it slightly increases the number of fundamentalist traders up to τ ≈ 0. But increasing the tax rate further crowds too many agents out of the market and thus leads to a falling tax income.3 0.5 dis(stock price) dis(output gap) 0.26 0. with respect to these variables.216 0.25 0.138 0.1% and decreases it gradually for higher tax rates.5 Tax Rate 1 0.5 Tax Rate 1 0.3%) the market is destabilized and the values of dis(s). Concerning stability. Increasing the tax rate (starting from zero). we evaluate the FTT by how well it is capable of reducing volatility and distortion of s.02 vol(output gap) 0. x and π with respect to the imposed FTT. leads to a sharp decline in the fraction of chartist traders which approximately equals zero for τ ≥ 0.18 0. x and π. dis(s).5 Tax Rate 1 4 Tax Revenue 3 2 1 0 0.1 0 0 x 10 -5 0.2 0.214 0.139 0.137 0.01 0. dis(x) and dis(π) recommend a diﬀerent conclusion.2 0.3%.23% leads to rising tax revenue.5 Tax Rate 1 0.24 0. Therefore. It stabilizes the market for small tax rates. Figure 7 shows the average fraction of chartists and fundamentalists.212 0. the FTT has an exclusively positive inﬂuence on stability.16 0.28 dis(inflation) 0 0.4 0.21 0. If it becomes too large (τ > 0. the FTT has an ambiguous impact.6%.22 vol(inflation) 0 0.218 0.005 0 0.

4. all gains from stock trading i (exp{st } − exp{st−1 }) Dt−2 are proﬁts. x and π fall. In our model. we do not consider labor costs in the ﬁnancial sector. Equation (7) changes to: i Ai = (exp{st } − exp{st−1 }) Dt−2 + dAi t t−1 i i − 1{R+ } (exp{st } − exp{st−1 }) Dt−2 · τ · (exp{st } − exp{st−1 }) Dt−2 (31) (32) We perform the same experiment that we used to analyze the impact of the FTT. Thus we introduce a FAT into our model by taxing proﬁts a constant rate of τ . IMF (2010) [p.28%.2 Financial Activities Tax A FAT. 69] and EU (2010) [p. To allow for a better comparability of the results. Several detailed examples of how such a FAT could look like can be found in the report (p. tax payments are given by: i i 1{R+ } (exp{st } − exp{st−1 }) Dt−2 · τ · (exp{st } − exp{st−1 }) Dt−2 Where 1{R+ } [y] is the indicator function that becomes 1 if y ∈ R+ and zero otherwise. For the FAT we assume diﬀerent values between 5% and 50%. For comparing both taxes with each other. All other measures follow a monotonic path. Results are illustrated in ﬁgure 8. 26 . 21). An optimal value of τ can not be identiﬁed. This rate leads to low market volatility and distortion at a high revenue (section 4. Tax revenues rise while volatility and distortion of s. we have to use a more stylized version of course. scaling of the ordinates is carried over from ﬁgure 7. 66-70). In our analysis we account for tax rates between 0% and 50%. but mainly one of political feasibility. The tax slowly decreases the number of chartist while keeping the number of fundamentalists almost constant. Increasing it leads to an improvement of all of our measures. The eﬀect of the FAT is thus clearly positive. 22. 6] assume a rate of 5%. If we assume further that the FAT only applies if proﬁts are positive. as proposed in the report of the IMF (2010) is ”levied on the sum of proﬁts and remuneration of ﬁnancial institutions” (p.38 The impact on all 38 For illustration purpose. The rate of a FAT has to be much higher than that of a FTT because its base is much smaller. Since. Therefore the question for the FAT’s size is not a question of economic optimality.1). we assume that the FTT is set in the range of optimal values at τ = 0.

8% for example).fundamentalists (solid) chartists (dashed) 0. is to identify a threshold value of proﬁts. Since the FTT leads to better (or at least equally good) results in all other respect.214 0.015 0.4 0. it is only a useful means to stabilize the market.212 0.26 0.01 20 40 Tax Rate 0.27 0. the FAT leads to a higher fraction of fundamentalists.005 0 20 40 Tax Rate 4 Tax Revenue 3 2 1 0 0. A point in favor of the FAT is that there is no danger of setting the rate too high. For all values. Adding such an element of progressivity. should discourage risk-taking.218 0.3 0 20 40 Tax Rate 0.136 0. The IMF also considers another variant of the FAT in its report and proposes: ”taxing high returns more heavily than low ” (p.22 vol(inflation) 0 20 40 Tax Rate 0. Note however. Let 27 .14 0.22 0.4 0.28 dis(inflation) 0 20 40 Tax Rate 0.2 0.02 vol(output gap) 0.16 0.24 0.137 0. Therefore. The ﬁrst step in deﬁning such a tax. dis(·) and vol(·)) are smaller.18 0.139 0. we perform the above experiment a third time.135 0 20 40 Tax Rate 0. huge distortions might occur (ﬁgure 7).2 0.25 0.5 vol(stock price) 0.216 0. All other measures are in favor of the FTT.1 0 0 x 10 -5 0. Instead of assuming a ﬂat tax as in equation (31) we let the tax rate grow with proﬁts. we can conclude that the FTT strictly dominates the FAT. If the optimal rate for the FTT is missed and a larger rate is set (0.2 0 20 40 Tax Rate 0 20 40 Tax Rate Figure 8: Impact of Financial Activities Tax (FAT) measures of interest are contrasted in table 2.3 0. that a large fraction of fundamentalists is not an end in itself. To test the scope of such a progressive FAT (called FAT3 in the IMF report).138 0.24 0. The tax revenue is larger or at least equally good for the FTT while all undesirable characteristics (W C .21 0. The FAT does not suﬀer from such a problem (ﬁgure 8) because of its monotonic impact.5 dis(stock price) dis(output gap) 0. Proﬁts above this threshold are deﬁned ’excess proﬁts’ and thus taxed higher. 68). it is only a measure of second order compared to values of distortion and volatility.

403 2. By increasing (decreasing) τ .400 a 10% 0.399 3.139 20% 0.173 0.404 2.273 0.228 0.015 0.212 0.139 0. We deﬁne the benchmark P ⋆ to be the standard deviation of proﬁts: P ⋆ := std(Pt ) A tax rate that is quadratically growing in P can be deﬁned as: FAT rate = 1{R+ } [P ] · τ · P P⋆ 2 (33) (34) Such a tax deﬁnition has the following nice properties.251 0. they are taxed by a rate of τ (ﬁgure 9).243 0.256 0.250 0.016 0.212 0.297 0. we restrict the tax to values ≤ 100%.6 ·10−5 0.0 ·10−5 0.402 1. we increase (decrease) the FAT for all (positive) values of P in the same direction while preserving the general shape.142 0. 28 .250 0.9 ·10−5 0.9 ·10−5 0.39 For proﬁts below the threshold. If proﬁts equal P ⋆ (the benchmark value).249 0.229 0.28% WC W F FAT 5% 0.140 Bold numbers indicate that the FAT leads to better results than the FTT i Pt = (exp{st } − exp{st−1 }) Dt−2 denote proﬁts.009 0.367 3. Figure 10 compares the results for the progressive FAT (bold line) with the ﬂat tax scenario of the 39 For very high proﬁts eq.234 0. It also allows us to perform experiments similar to the ones above.2 ·10−5 0. Since such rates are unrealistic.139 a Revenue dis (s) dis (x) dis (π) vol (s) vol (x) vol (π) 0.211 0. If they are above P ⋆ (excess proﬁts) they are taxed by a higher rate that grows quadratically in P .139 50% 0. The tax rate is thus progressive and not subject to any steps.174 0.139 35% 0.249 0.171 0.172 0.211 0.249 0.Table 2: Comparison of FTT with FAT FTT 0.5 ·10−5 0.171 0.054 0.014 0.212 0. (34) might result in tax rates above 100%.250 0.012 0. the tax falls and smoothly approaches zero at P = 0.173 0.179 0.212 0.011 0.231 0.

The progressive FAT gives rise to a better stabilization for rates up to τ ≈ 30%: The number of fundamentalists is higher while the number of chartists is lower compared to the ﬂat FAT.FAT rate τ 0 0 Profits P* Figure 9: Rate of the progressive FAT previous experiment (thin line).2116 vol(inflation) 0. Only for high tax rates of 30% or more the results turn around partly: vol(s).1396 0.4 0.171 2 0.012 0.14 0.2112 flat 20 40 Tax Rate 0.248 0. fundamentalists (solid) chartists (dashed) 0. vol(x). The revenue is larger under a progressive FAT between 0% and 11%.1 0.175 0.26 0.01 0 20 40 Tax Rate 0. At the same time.2114 0.016 0. The nice property of monotonically decaying volatility and distortion that has already been found for the ﬂat FAT is also found for the progressive one.25 0.23 0.014 0.17 0 20 40 Tax Rate Figure 10: Impact of progressive Financial Activities Tax (FAT) The reason for this result can easily be explained by taking a look at the distribution of proﬁts.174 0.018 0.173 0.139 0 20 40 Tax Rate 3 Tax Revenue 0.22 0 20 40 Tax Rate 0.251 dis(inflation) 0 20 40 Tax Rate 0. dis(x) and dis(π) are advantages under a ﬂat tax.3 0. Comparing both FAT’s leads to ambitious results.24 0.1394 0.211 0 20 40 Tax Rate 0.27 dis(stock price) dis(output gap) 0.172 0. all volatility and distortion measures are lower. For tax rates above 11% the ﬂat FAT yields more income. Figure 11 shows the distribution of fundamentalist’s proﬁts (solid line) and chartist’s proﬁts (dashed 29 .5 vol(stock price) 0.02 vol(output gap) 0.2 0.249 0 0 20 40 Tax Rate 0.1392 progressive 0 x 10 -5 0. dis(s).25 1 0.1398 0.

in a semi-log scaled plot and a log-log scaled plot. 40 Consult Lux (2009) on the empirical properties of ﬁnancial data. For all values of the tax.28%.e. 30 . If we leave the fractions of fundamentalists and chartists aside. Fundamentalists (who earn lower proﬁts on average) are thus favored by the progressive FAT. for very high proﬁts) both distributions are approximately equal. the fraction of fundamentalists increases while that of chartists decreases compared to a ﬂat FAT (ﬁgure 10. We can therefore conclude that the classical FTT is better suited than the new FAT in order to achive stabilization of markets and raise funds from the ﬁnancial sector. As a result. line). the distribution of fundamentalist’s proﬁts has more density located at low values while chartists have more density located at higher values.40 At the same time. upper left panel). Both lines show the characteristic fat tails. both. Results are given in table 3. the FTT again strictly dominates the FAT. In the tail (i.2000 Fundamentalists Chartists 10 4 1500 Probability 10 Probability 2 Fundamentalists Chartists 1000 10 0 500 10 −2 0 10 −4 10 −3 −2 −1 0 −4 10 10 Profits 10 10 10 −4 10 −3 10 Profits −2 10 −1 10 0 Figure 11: Distribution of Proﬁts in a semi-log (left) and log-log scaled plot (right). the fraction of fundamentalists is higher than under the FTT while the values of vol (π) are equal throughout all simulations. A more stable economy is the result. while chartists (who earn higher proﬁts on average) are disadvantaged. Finally. we compare the progressive FAT at diﬀerent rates with the FTT of 0.

262 0.212 0.449 1.28% 3.243 0. 5.209 0.139 20% 0.7 ·10−5 0.249 0.203 0.226 0.171 0.28% WC W F FAT (progressive) 5% 0.460 1.1 Parameterization As mentioned in section 3. we check the robustness of our result with respect to some assumptions that we had to make throughout our analysis.367 3.012 0.178 0. Table 4 and 5 show how some important results change with the variation of φ and e.233 0.26% φ = 15 0.2 ·10−5 0.139 a Revenue dis (s) dis (x) dis (π) vol (s) vol (x) vol (π) 1.228 0.139 35% 0.139 50% 0.28% 3.28% 0.250 0.6 ·10−5 0.29% 3.Table 3: Comparison of FTT with progressive FAT FTT 0.211 0.28% 0.30% 3.235 0.205 0.214 0.2 ·10−5 0.172 0. we focus on the robustness of our derived optimal FTT.234 0.212 0.233 0.2 ·10−5 0.2 ·10−5 0.8 ·10−5 0.466 1.2 ·10−5 0.239 0.249 0.26% φ = 10 0.30% 0.241 0.212 0.2 ·10−5 0.258 0.6 ·10−5 0.28% 0.26% φ=8 0.212 0.7 ·10−5 0.009 0.014 0.26% φ = 12 0.212 0. calibration of the intensity of choice parameters φ and e is probably the most problematic one.172 0.25% 0.424 a 10% 0.139 0.212 0.436 1.172 0.26% 31 .054 0.012 0.172 0.013 0.249 0.27% 3.139 Bold numbers indicate that the FAT leads to better results than the FTT 5 Robustnes Checks In this section.249 0.015 0.162 0.255 0.211 0.249 0. Since our simulations suggests that the FTT is the best way of taxing the ﬁnancial sector.171 0. Table 4: Robustness check of intensity of choice parameter φ φ=5 minτ vol(x) arg minτ vol(x) minτ dis(x) arg minτ dis(x) minτ Revenue arg minτ Revenue 0.249 0. Our ﬁrst robustness checks are therefore concerned with these parameters.

81% e = 200 0.80% 8. The negative eﬀects of a too low FTT will instead be much smaller.28% 3.4 ·10−5 0.26% e = 400 0.18% e = 500 0.211 0.54% 0.211 0. We have shown in section 4.3 ·10−5 0.25% 0.Our results are fairly robust against diﬀerent values of the parameter φ (table 4).211 0.4 ·10−5 0.41% 4. The expectations and fractions of optimistic and pessimistic central 32 .28% 0.7 ·10−5 0. The induced Tax revenues are also subject to high uncertainty. The optimal tax rate diverges strongly between 0. Table 5: Robustness check of intensity of choice parameter e e = 100 minτ vol(x) arg minτ vol(x) minτ dis(x) arg minτ dis(x) minτ Revenue arg minτ Revenue 0.36% 0. we assume that the board of the central bank is composed of a set of heterogeneous agents who form expectations diﬀerently. We begin by introducing an interest smoothing parameter into (9).250 0. Variation A of the policy rule is thus given by: ˜ ˜ iq = λ δπ Eq [πq+1 ] + δx Eq [xq+1 ] + (1 − λ)iq−1 + ǫi q (35) As a second variation. that reliable estimations of the intensity of choice parameter for the ﬁnancial sector are of major importance for reliable policy suggestions. A good strategy might thus be to set the FTT signiﬁcantly below the value that is optimal with respect to a given parameterization in order to deal with the uncertainty in an appropriately careful way.251 0.16% 5.248 0.41% e = 300 0.211 0.12% 2. This robustness test suggests. dis(x) and Revenue change only slightly.18% 0.10% 2.248 0.2 ·10−5 0.249 0. Some are optimistic about future output and some are pessimistic.54% for diﬀerent values of e (table 5). The optimal tax rate as well as the induced optimal values of vol(x).18% and 0.211 0.1 that a too high FTT can result in huge distortion of the real economy. The current interest rate is thus not only concerned with reducing inﬂation and the output gap but also with producing a smooth path of iq over time.2 Taylor Rule In this section we test the robustness against diﬀerent speciﬁcations of the Taylor rule. The opposite holds for the parameter e.

08% for variation B.3 ·10−5 0. It falls by 0. 33 .27% (37) Table 6 compares our result for the diﬀerent speciﬁcations of the Taylor rule.28% 0. The optimal tax rate that minimizes vol(x) undergoes the largest change.249 0. The expectation that enters the Taylor rule is then given by: ˜ opt Eq [xq+1 ] ˜ pes E [x q q+1 ] pes opt if ωq > ωq ˆ Eq [xq+1 ] = (36) otherwise In other words.2.20% 0.opt ˜ opt ˜ pes bankers are the same as for the other agents.272 0. ωq pes and ωq as deﬁned in section 2. the central bank’s decision structure is such that it become fully optimistic if optimists dominate and fully pessimistic if pessimists dominate.278 0.41 The results show that the measures of stability and tax revenue are only subject to minor change.26% Variation A (λ = 0.203 0.29% 3. Variation B of the Taylor rule is given by: ˆ ˆ iq = δπ Eq [πq+1 ] + δx Eq [xq+1 ] + ǫi q ˆ ˆ Where Eq [πq+1 ] is formed analogous to Eq [xq+1 ].28% 3. Table 6: Robustness check of diﬀerent versions of the Taylor rule Taylor Rule minτ vol(x) arg minτ vol(x) minτ dis(x) arg minτ dis(x) minτ Revenue arg minτ Revenue Baseline speciﬁcation 0. they are given by Eq [xq+1 ].24% 0.04% in case of variation A and 0.75.228 0.26% Variation B 0. Assume that the majority of central bankers can fully introduce their expectations into the policy rule. Thus.211 0. The parameter λ in Variation A is set to λ = 0. Eq [xq+1 ].24% 3.2 ·10−5 0. 41 The same analysis for a wider λ-range as well as the impulse response functions resulting from the variation of the Taylor Rules can be found in the online appendix.75) 0.2 ·10−5 0. This uncertainty again suggests that the rate of the FTT should be set to lower rates compared to those that are optimal given our baseline speciﬁcation.

The two employed submodels are simple representatives of their respective discipline. we are able to show that the behavioral structure of our model has a strong impact on the transmission of shocks. Our research agenda is targeted at further 34 . while for rates above 40% the ﬂat tax version is preferable. But in contrast to the FTT. The FAT is less able to stabilize the market and also generates less revenue for the state. if the introduction of a tax on ﬁnancial transactions can bring about positive developments for the overall economy. In between. We have shown that the optimal decision of making the FAT ﬂat or progressive is depending on the tax rate. If the tax is of the FTT type. Interaction between the two models is brought about by two straightforward channels. Tobin’s q or stock wealth eﬀect) can be analyzed. (2) The rules that deﬁne the behavior of the ﬁnancial market agents (like the time horizon of investors’ strategies) can be changed. For values below 11% the progressive version is the best choice. (4) Moreover. for example. we do not take ﬁnancial streams between the real and the ﬁnancial sector explicitly into account. it is very eﬃcient in bringing down volatility and raising tax revenue. (3) Since the occurrence of bubbles implies large deviations from the fundamental steady state. one might also use a version of the NKM submodel that is not log-linearized. We ﬁnd that such a tax could generally reduce volatility and distortion of the real and ﬁnancial market variables. but if set too high. but that its size and type plays an important role. can lead to a very diﬀerent average transmissions. the progressive tax leads to better stabilization while the ﬂat tax generates more revenue. But even with this simplistic methodology. We see this paper as an early stage in a broader research agenda.g.6 Conclusion We have developed a model that combines agent-based ﬁnancial market theory with New Keynesian macroeconomics. We also used the model to analyze a question that is currently debated among policy makers. Our comprehensive model is very stylized and not yet ready for econometric analysis. it does not create large distortions when set too high. the macroeconomy might also be subject to very strong distortion. The market sentiments in the shock period. it can also be used for numerous augmentations: (1) The eﬀects of diﬀerent cross-sectoral channels (e. Of course. The agent-based method oﬀers enormous new possibilities for macroeconomics. Namely. A simpliﬁcation that might be relaxed in future research. Our model is stylized and simple to implement. They are both subject to an evolutionary process of expectation formation that sorts out the poorly performing strategies in favor of the good ones.

Saarland o University. F. Journal of Economic Dynamics and Control 22: 1235–1274. M. The Econometrics of Financial Markets. B. 42 See for example Gaﬀeo et al. W. Bask. C. Working Papers 540. Hanken School of Economics. Princeton University Press. chapter 21. References Ackerman. Brock. (1998). It uses new methods (like interaction and evolutionary learning). (1997). G. Reconstructing Macroeconomics: A Perspective from Statistical Physics and Combinatorial Stochastics Processes. (2002). Porter. A. Friedmann. Aoki. Caginalp. (2007). Stock Price Misalignments and Macroeconomic Instability. A. and Hommes. R. Working Papers. M. (2008) or Dosi et al. The paper at hand tries to bridge the gap between this newly emerging ﬁeld42 and mainstream macro. and Nistico. V. Stock Market Conditions and Monetary Policy in a DSGE Model. In: Handbook of Macroeconomics. (2010). H. Momentum and Incomplete Information. Still Dead After All These Years: Interpreting the Failure of General Equilibrium Theory.. Volume I. Y. and Gilchrist. Woodford.. pp. S. C. and Yoshikawa. Gertler. Monetary Policy and Asset Price Volatility.. The Financial Accelerator in a Quantitative Business Cycle Framework. Bernanke. B. E. but also builds on traditional methods and assumption (like market equilibrium or utility maximization in the NKM part). Campbell. (2008). W. Economic Review 4: 17–51. and MacKinlay. Taylor and M. Journal of Economic Dynamics and Control 34: 1700–1731.exploration of these possibilities. and Smith. (2007). Journal of Economic Methodology 9(2): p. Lo. Cambridge University Press. Intraday Overreaction of Stock Prices. Financial Bubbles: Excess Cash. and Gertler. S. H. Future research will focus on working out the agent-based part further. Becker. (1999). B. and Kl¨ßner. Bernanke. 35 . S. D. Journal of Behavioral Finance 2: 80–99. Castelnuovo. 1341–1393. S.. J. Heterogeneous Beliefs and Routes to Chaos in a Simple Asset Pricing Model. Monetary Policy. (2009). M. A. M. Edited by J. M. (1999). 119–139. (2001).

Press Releases IP/11/799. Fagiolo. The Financial Crisis and the Systemic Failure of the Economics Profession. Does the Stock Market Overreact?. D.. Agent-Based Models for Economic Policy Design. and Sloth. (2010b). G. (2006). H. D. M. American Economic Review 98: 236–240. (forthcoming). (1985). Colander. Haas. Demary. M.. M. o B. European Commission Services: Working Document SEC(2010)1166 .. and Gallegati. Economics: The Open-Access. Delli Gatti. A. Jahrb¨cher f¨r National¨konomie und Statistik (Journal of Economics u u o and Statistics) 228: 228–250. Transaction Taxes and Traders with Heterogeneous Investment Horizons in an Agent-Based Financial Market Model. Exchange Rate Puzzles: A Tale of Switching Attractors. and Dawid. E. Journal of Economic Interaction and Coordination. EU (2010). G. Animal Spirits and Monetary Policy. Howitt. Applied Mathematics and Computation 204: 541–552. M. (2008). T. Public Choice 144: 413–443. F¨llmer. (2009). 249–267. M.. Eastern Economic Journal. W. P. Beyond DSGE Models: Toward an Empirically Based Macroeconomics. (2008). (2010a). The Journal of Finance 40: 793–805. (2008). De Bondt. Kirman. Kirman. F.. P. and Grimaldi. and Neugart.. H. EURACE: A Massively Parallel AgentBased Model of the European Economy. R. Dawid. Multi-Agent Model. K. A. Journal of Evolutionary Economics 18: 413–432. European Economic Review 50: 1–33. C. De Grauwe. Dosi. A. EU (2011a).. and Thaler.. 36 . Gaﬀeo. The Microfoundations of Business Cycles: An Evolutionary. Demary. Complex Agent-Based Macroeconomics: A Manifesto for a New Paradigm. and Mehrling. and Roventini. P. Juselius. A. Critical Review 21(2-3): p.. Who Does a Currency Transaction Tax Harm More: Short-Term Speculators or Long-Term Investors?. Goldberg.. M. Open-Assessment E-Journal 4.. (2010). van der Hoog.. De Grauwe. M. (2010). P. Economic Theory. The scientiﬁc foundation of dynamic stochastic general equilibrium (dsge) models. (2008). Lux. Investing Today for Growth Tomorrow. Deissenberg. D. A. P. Leijonhufvud. H. De Grauwe. Financial Sector Taxation.Colander. S.

and van de Velden. In: Handbook of Computational Economics. Desiderio. (2009). Using Survey Data to Test Standard Propositions Regarding Exchange Rate Expectations. (2008). Economic Working Papers 1. How Did Economists Get it so Wrong?. Journal of Macroeconomics 10. pp. The American Economic Review 80: 434–449. Has Macro Progressed?. chapter 23. Coordination of Expectations in Asset Pricing Experiments. The American Economic Review 77: 133–153. Scottish Journal of Political Economy 53: 636–654. Economic Modelling 22: 1105–1121. Foreign Exchange Rate Expectations: Micro Survey Data. (2010).. The B. Keen. CESifo Economic Studies 57. 108: 137–156. Judd. Hommes. (2006). P. Staﬀ of the International Monetary Fund: Final Report for the G-20 . Kirman. (1987). Eastern Economic Journal 34: 441–463. A. Kontonikas. H. J. A Fair and Substantial Contribution by the Financial Sector.EU (2011b). S. C. and Montagnoli. (2005). The Quarterly Journal of Economics 37 . Gaﬀeo. Sonnemans. Volume 2.. Should Monetary Policy Respond to Asset Price Misalignments?. (2010). C. C. Kontonikas. Krugman. A. T. M. IMF (2010). (2011). (2006). and Ioannidis. M... A. Ito. Kirman. J. H. (2009). A. Press Releases MEMO/11/468. Cowles Foundation Discussion Papers 1728. A. (1990).E. and Froot. (1993). A. Franke. Lam. J. The Economic Crisis is a Crisis for Economic Theory. Rethinking the Taxation of the Financial Sector. Christian-Albrechts-Universit¨t Kiel. Tuinstra. K. D. The Importance of Commitment in the New Keynesian Model. New York Times Sept 2.. 1009–1186. A. Rationality and Recruitment. Some Observations in the High-Frequency Versions of a Standard New-Keynesian Model. Optimal Monetary Policy and Asset Price Misalignments. (2005). a Frankel. R. Hommes. Ants. C.-P. R. and Sacht. and Gallegati. Review of Financial Studies 18: 955–980. Adaptive Microfoundations for Emergent Macroeconomics. Heterogeneous Agent Models in Economics and Finance. (2010). Multiannual Financial Framework (MFF): Questions and Answers. S. Fair. Edited by Leigh Tesfatsion and Kenneth L. J. Delli Gatti. E. Paper presented at the CESifo Economic Studies Conference on “What’s Wrong with Modern Macroeconomics?”.

38 . and Avard. 1187–1233. S. D. (2000). B. Judd. Microscopic Models for Financial Markets. and Westerhoﬀ. (2005).. Manski. Learning About the Interdependence Between the Macroeconomy and the Stock Market. Zschischang. E. G. chapter 24. Taxing Financial Transactions: Issues and Evidence. chapter 3. Cambridge. Exchange Rate Determination. In: Handbook of Computational Economics. Feasibility and Eﬀects. Department of Economics Working Papers 070819. Journal of Monetary Economics 53: 199–216. (2006). Journal of Economic Behavior & Organization 77: 177–188. Lux. Revenues. C. (2008).. International Journal of Theoretical and Applied Finance 3: 675–702. (2008). T. M. Journal of International Money and Finance 17: 535– 545. Y. and Picek. The Use of Fundamental and Technical Analyses by Foreign Exchange Dealers: Hong Kong Evidence. and Marchesi. D. K. Asymmetric Reverting Behavior of Short-Horizon Stock Returns: An Evidence of Stock Market Overreaction. Monetary Policy and Asset Price Bubbles. T. Lux. SchenkHopp´. C.-H. Rudebusch. Proa˜o. (1981). (2006). MIT Press. (2011). e Lux. E. C. Volatility Clustering in Financial Markets: A Micro-Simulation of Interactive Agents. Osterreichisches Institut f¨r Wirtschafsforschung u Working Paper.. R. Optimal Monetary Policy with the Cost Channel. Agent-Based Computational Finance. O. Economic Working Papers 15. E. Macroeconomic Dynamics and Stability unn der Heterogeneous Behavioral FX Expectations. Christian-Albrechts-Universit¨t Kiel. Pyun. L. Edited by Leigh Tesfatsion ans Kenneth L. a Schulmeister. Journal of Banking & Finance 25: 807–824. Stauﬀer. IMF Working Paper 54. C. Stochastic Behavioral Asset-Pricing Models and the Stylized Facts. (2006). (2009). University of California-Irvine.. Nature Physics 5. T. Edited by Thorsten Hens and Klaus R. A General Financial Transaction ¨ Tax: Motives. S. Structural Analysis of Discrete Data with Econometric Applications. and Walsh. M. Ravenna. 161–215. F. (1998). pp. F. and McFadden. and Lux. pp. Matheson. S. Samanidou. and Mole.LeBaron. Lui. F. Schratzenstaller. (2001). Volume 2. Federal Reserve Bank of San Francisco Economic Letter 5. Economics Crisis. (2011).. Milani. In: Handbook of Financial Markets: Dynamics and Evolution. Nam. T. T. (2009). D. D.

and Dieci. Hommes. Taylor. Tuinstra. Jahrb¨cher f¨r National¨konomie und Statistik (Journal of Economics and u u o Statistics) 228: 195–227. Discretion versus Policy Rules in Practice. H. Heterogeneous Traders and the Tobin Tax. (1992). M. (2004). and van de Velden. Westerhoﬀ.Sonnemans. Journal of Economic Dynamics and Control 30: 293–322. Carnegie-Rochester Conference Series on Public Policy 39: 195–214. (2008). 39 . Journal of International Money and Finance 11: 304–314. P. Journal of Economic Behavior and Organization 54: 453–481. F.. J.. C. Journal of Evolutionary Economics 13: 53–70. (2003). B. H. Westerhoﬀ. J. J. R. (1993). (2006). Westerhoﬀ. (1978). Eastern Economic Journal 4: 153– 159. and Allen. H. The Eﬀectiveness of Keynes-Tobin Transaction Taxes When Heterogeneous Agents Can Trade in Diﬀerent Markets: A Behavioral Finance Approach. A Proposal for International Monetary Reform. The Use of Agent-Based Financial Market Models to Test the Eﬀectiveness of Regulatory Policies. F. J. Taylor. F. The Instability of a Heterogeneous Cobweb Economy: A Strategy Experiment on Expectation Formation. Tobin. The Use of Technical Analysis in the Foreign Exchange Market.

2011 ∗ ∗∗ E-mail: matthias.uni-mannheim.de E-mail: wohltmann@economics.Online Appendix Agent-Based Financial Markets and New Keynesian Macroeconomics – A Synthesis – Matthias Lengnick∗ University of Kiel University of Mannheim Hans-Werner Wohltmann∗∗ University of Kiel September 13.uni-kiel.lengnick@vwl.de .

211 0.28% 0. Equation (24) becomes: 64q sq = t=64(q−1)+1 weightt st 0<ρ<1 (1) (2) with: weightt = (1 − ρ) · ρ64q−t High values for ρ lead to slowly decaying weights.251 0.28% 0.211 0. For a smaller period of 64 days they do not.249 0.2 ·10−5 0.26% 0.2 ·10−5 0.9 0. Geometric weights of this form add up to one for very long time periods.249 0.28% 0.211 0. 1 .251 0.28% 0.28% 0.28% 3.and ﬁnancial sector.211 0.75 0.26% Table 1 shows that our results are very robust against the assumption of diﬀerent weights in (24).28% 3.5 0.28% 3. An alternative would be to assume geometrically decaying weights.2 ·10−5 0.26% ρ = 0.26% ρ = 0.26% geometrically decaying ρ = 0.99 minτ vol(x) arg minτ vol(x) minτ dis(x) arg minτ dis(x) minτ Revenue arg minτ Revenue 0. The minimal distortion of the output gap changes marginally.1 Robustness to Diﬀerent Deﬁnitions of Weights For integrating the real. low values result in quickly decaying weights. we have assumed that all daily realizations of st enter the quarterly value sq with equal weights. We thus rescale our weights by multiplying with a constant that keeps the relative weight between diﬀerent st constant but yields 64q weightt = 1 t=64(q−1)+1 (3) Table 1: Robustness check of weighting assumption in equation (24) Weights equal ρ = 0.28% 3.2 ·10−5 0.211 0.28% 3.2 ·10−5 0.250 0. All other results stay constant.

1 Taylor Rule Variation A of the Taylor rule has been deﬁned as: ˆ ˆ iq = λ δπ Eq [πq+1 ] + δx Eq [xq+1 ] + (1 − λ)iq−1 + ǫi q (4) Table 2 shows the results of a robustness check for diﬀerent values of λ.355 0. Table 2: Robustness check of the interest smoothing parameter λ λ = 1.28% 0.278 0.27% 0.28% 3.249 0.11% 3.08% 3.2 ·10−5 0.2 ·10−5 0.3 ·10−5 0.5 ·10−5 0. 2 . Moving away from our baseline calibration (λ = 1.9 0. If λ becomes too small.543 0. The Taylor rule depends on expectations about the future values of x and π.75 0.257 0.203 0.75 and rises afterwards.0) monotonically increases distortion of x in the optimum.26% λ = 0.18% Concerning the tax rates.26% λ = 0.235 0. A moderate weight on interest smoothing therefore reduces volatility of output while a high weight leads to an increase. the inﬂuence of the stabilizing role of monetary policy declines by so much that vol(x) rises again.211 0. Reducing the weight of these expectations therefore leads to a decline of vol(x).24% 3. instead.1. we ﬁnd again that an optimal value might lie signiﬁcantly below the values that have been found to be optimal in our baseline calibration.212 0.5 0.27% 3.14% 0.0 minτ vol(x) arg minτ vol(x) minτ dis(x) arg minτ dis(x) minτ Revenue arg minτ Revenue 0. The volatility of x.3 0. they are a potential source of volatility.206 0. Since these are formed in a non-rational way. falls until λ = 0.2 ·10−5 0.26% λ = 0. The reason for this result is the following.24% 0.22% 0.25% λ = 0. Figures 1 and 2 show the impulse responses under our alternative Taylor rules. This result obviously follows from the fact that the central bank is less concerned with stabilizing output and more with smoothing the interest rate.

3 q 3 Real Interest Rate .5 Output Gap 0 −0.5 0 Stock Price 10 20 30 0.5 0 −0.5 −1 0 1 Inflation 0.5 0 −0.5 −1 0 10 20 30 10 20 30 10 20 30 ˆ ˆ Figure 1: Cost shock for iq = δπ Eq [πq+1 ] + δx Eq [xq+1 ] + ǫi q 2 Output Gap Stock Price 10 20 30 1 0 −1 −2 0 1.5 0 −0.5 0 −0.5 Inflation 1 0.5 −1 0 10 20 30 10 20 30 ˆ ˆ Figure 2: Cost shock for iq = λ δπ Eq [πq+1 ] + δx Eq [xq+1 ] + (1 − λ)iq−1 + ǫi with λ = 0.0.5 0 −0.5 −1 0 1 0.5 0 10 20 30 Real Interest Rate 2 1 0 −1 −2 0 1 0.

- Andreou Pitt is s Panos 2001
- PURCHASE ONLYInstruire Loc Munca1
- Instruire Loc Munca1
- Instruire generala1
- 15th Global Ceo Survey
- 16th Global Pwc Ceo Survey
- 15th Global Ceo Survey
- Cavalerul Medieval
- Starea Micilor
- Barman(1)
- Einstein.quest
- TM3W Rate Card
- Registru evidenta decizii
- Goog Graph (2)
- Calcul Stochastic_curs Introductiv II_slide
- Normal (1)
- EVALUARE DULGHER 1
- Responses to Criticisms FTT
- TEST SSM Angajare
- Plan de Evacuare
- Cum Sa Tranzactionam Pe Bursa

- gewb
- wp13-36
- Figlewski Stephen - What Does an Option Pricing Model Tell Us About Option Prices
- Subscribe Rss Feeds
- Stock Price Formation From a Value Standpoint
- Fix income trading strategy
- Sornette_HFT
- Ch12 Management Compensation
- Investor eBook
- Movement for Global Cooperation - The Bio-compatible Economy, a Summary
- Chaos Theory
- Nature, Scope and Significance of Maangerial Economics.pptx
- Possible Vehicle Document Verification Portal Network Revenue Stream1
- August 2010 - Defined Benefit Pension Plans
- 19538242 Technical Analysis Notes
- IB0010–International Financial Management
- 52-13239-1-PB (1)
- out5.pdf
- rev_frbsf_196406
- Intellectual Capital
- Engineering Notes
- Amit Kumar Verma Pgd11007 x Ks (2)
- Daily MCX Tips and Newsletter 23April2013.pdf
- Globalstar GSAT Report
- Newsletter Banca Italia
- BPO - India a perspective
- Marketing Renaissance_steve Burgess
- New Microsoft Word Document (2)
- Hopkins (2)

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue listening from where you left off, or restart the preview.

scribd