Agent Based-Stock Flow Consistent Macroeconomics: Towards a

Benchmark Model
Alessandro Caiani

Antoine Godin

Eugenio Caverzasi

Marche Polytechnic University

University of Limerick

Marche Polytechnic University

Mauro Gallegati

Stephen Kinsella

Joseph E. Stiglitz

Marche Polytechnic University

University of Limerick

Columbia University

June 25, 2015

Abstract
The global financial crisis has forced standard macroeconomics to re-examine the plausibility of its
assumptions and the adequacy of the policy prescriptions flowing from those assumptions. We believe
a renewal of macroeconomic thinking and macroeconomic modeling is possible by recognizing that our
economies should be analyzed as complex adaptive systems. A coherent and exhaustive representation
of the inter-linkages between the real and financial sides of the economy is vital as well. We propose
a macroeconomic framework based on a novel combination of the Agent Based and Stock Flow
Consistent approaches. This paper presents a benchmark model for this innovative approach. Our
model depicts an economy with capital and credit in which different types of agents locally interact on
different markets. We provide a detailed representation of individual agents balance sheets, ensuring
the model accounting consistency at the micro, meso, and macro levels. We analyze the properties
of our simulated economy under different configurations of agent heuristics, focusing in particular on
the role of credit and investment. We explain in detail the logic followed to calibrate and validate the
model. Results show that our benchmark model is able to reproduce many stylized facts observed in
real world, thus representing a good starting point to test - in the next works - different economic
policies and institutional setups. Finally, the relatively simple and flexible structure of the model
opens up many possibilities for development of the framework along different lines, thus providing a
fertile soil for new applications.

Keywords: Agent Based Macroeconomics, Stock Flow Consistent Models, Business Cycles, Crisis.
JEL Codes: E03, E32, O30

Corresponding author: [email protected]. This research was supported by the Institute for New economic Thinking
(INET) and the FP7 project MatheMACS. Our work has greatly benefited from comments and suggestions received from
other scholars. We are grateful to the participants of the 2014 Workshop of the INET AB-SFC Macroeconomic Program
at Monte Conero. A special thanks goes to Steve Phelps for the support he gave us while developing the JMAB platform.
We thank our colleagues Ermanno Catullo, Annarita Colasante, Federico Giri, Ruggero Grilli, Antonio Palestrini, Luca
Riccetti, Alberto Russo, Gabriele Tedeschi, and Sean Ryan who provided insight and expertise that greatly assisted the
research. All remaining errors are ours.

Contents
1 Introduction

2 Related literature

3 The model
3.1 Dispersed interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2 Sequence of events . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

6
7
7

4 Agent behaviors
4.1 Firm behavior . . . . . . . . . . . . . . . . . . .
4.1.1 Production planning and labor demand
4.1.2 Pricing . . . . . . . . . . . . . . . . . .
4.1.3 Firms profits . . . . . . . . . . . . . . .
4.1.4 Investment . . . . . . . . . . . . . . . .
4.1.5 Firms finance . . . . . . . . . . . . . . .
4.1.6 Labor, Goods and Deposit markets . . .
4.2 Bank behavior . . . . . . . . . . . . . . . . . .
4.3 Firms and banks bankruptcy . . . . . . . . . .
4.4 Household behavior . . . . . . . . . . . . . . . .
4.5 Government and central bank behavior . . . . .

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

5 Baseline setup: challenges in calibration

6 Results
6.1 Validation . . . . . . . . . . . .
6.2 The transition . . . . . . . . . .
6.3 Sensitivity analysis . . . . . . .
6.3.1 Differential risk aversion
6.3.2 Investment behavior . .
6.3.3 Credit Demand . . . . .

. . . .
. . . .
. . . .
among
. . . .
. . . .

8
8
8
9
9
10
10
11
11
14
15
15
16

. . . .
. . . .
. . . .
banks
. . . .
. . . .

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

18
18
20
24
24
27
28

7 Conclusions

28

A Parameters values and Initial Setup

32

B Auto and cross correlations in our experiments

33

Introduction

More than seven years since the onset of the global financial crisis, we are still assessing how the crisis
should change our views about macroeconomic policy. The crisis has cast doubt on the ability of standard
macroeconomic models - in particular of Real Business Cycle and New-Keynesian dynamic stochastic general equilibrium (DSGE) models - to explain the functioning of our economic systems and, perhaps more
importantly, to provide adequate advice on appropriate stabilization policies to prevent the occurrence
of large-scale economic turmoil and tackle its consequences. As pointed out by (Blanchard et al., 2012,
p.57):
The workhorse New Keynesian dynamic stochastic general equilibrium (DSGE) models on
which we were concentrating so much of our attention have been of minimal value in addressing
the greatest macroeconomic crisis in three-quarters of a century.
The DSGE community responded to the crisis1 with new models embedding some types of financial
frictions. Despite this, we feel that standard macroeconomics is still far from having taken the lessons of
the crisis on board. Our contribution goes beyond mainstream models in developing a relatively simple,
general, and flexible model based on the following conceptual pillars.
Micro- and macro-economics are different. The representative agent approach is inherently affected
by the fallacy of composition in assuming that what is true for individual agents is also valid for the whole
economic system. We believe on the contrary the correct way of linking the micro, meso, and macro layers
of the economy should be geared around the concept of complex adaptive systems, showing independent
system-wide properties emerging from agents disperse interactions (Delli Gatti et al., 2010a).
Agents are not fully rational nor fully informed. The decision-making process by agents is characterized by the prevalence of satisficing behaviors over optimizing ones, due to the presence of informational and computational limitations (Simon, 1955, 1976). Adaptivity and heterogeneity of agents
beliefs and heuristics should be a key ingredient to evaluate the effects of alternative policy formulations
(Anufriev et al., 2013).
The real and financial sides of the economy are closely interrelated. Every macroeconomic model
designed to perform fiscal and monetary policy analysis cannot abstract from a proper representation of
the financial system and the process by which inside money (i.e. money issued by private intermediaries
in the form of debt) and outside money (i.e. government-issued money) are created, injected in the
monetary circuit, and destroyed.
Every macroeconomic model should provide a complete and coherent accounting system at both the
individual and aggregate levels. Economic agents are connected to each other through the stocks they
hold in their balance sheets, either as assets or as liabilities. Decisions undertaken by individual agents
and resulting in a variation of their balance sheet affect other agents balance sheets both directly and
indirectly. In order to track these effects and avoid having black holes in the accounting framework, every
financial stock should then be recorded as an asset for someone and a liability for someone else. Similarly,
every flow should be an inflow for someone and an outflow for someone else.
In order to embed these features in our model, we combine the bottom-up perspective characterizing
the Agent Based (AB) modeling approach (Farmer and Foley, 2009; Esptein, 2006), with the rigorous
accounting framework which we find at the base of the Stock-Flow Consistent (SFC) framework (Godley
and Lavoie, 2007).
In this way we aim to provide a rigorous and realistic benchmark model to assess the effects of
different economic policies. The rigor of our analysis requires some preliminary steps to be made before
the model can be used for this purpose. These steps are aimed at improving our understanding of the
functioning of the model, identifying the underlying dynamics, and investigating its properties under
different circumstances.
In the present work we provide a detailed discussion of the logic followed to calibrate the model
parameters and set the initial values of stocks and flows in a consistent way; we analyze the dynamics
of the model under the baseline scenario; we compare the properties of our artificial time series with
real world ones to assess whether the model provides an acceptable approximation of reality; finally, we
perform sensitivity experiments on some relevant behavioral parameters to check the robustness of our
results and to investigate the properties of our simulated economy as a result of different behavioral
specifications.
Our artificial time series are analyzed by separating the cyclical and trend components. The model
shows how heterogeneity endogenously emerges, starting from initial symmetric conditions, leading to
1 See

for example Borogan Aruoba et al. (2013); Andreasen et al. (2013); Andreasen (2013); Benes et al. (2014)

interesting dynamics and changing market structures. Our results show that the model generates persistent economic fluctuations whose properties are similar to those observed in real world data. The model
is also capable of replicating several other stylized facts concerning, for example, the distribution of firms
size and banks credit degree distribution. The analysis of the long term dynamics highlights that, under
the majority of cases analyzed, the economy tends to converge to what we defined as a quasi steadystate, that is a temporarily stable configuration in which real aggregates trends are fluctuating around
a steady level, and nominal variables growth rates fluctuates around constant values, and inter-sectoral
flows and stock-flow ratios tend to stabilize. These outcomes will be the starting point to perform policy
and scenario analysis in further publications.
The rest of the paper is organized as follows: in the next section we briefly discuss the literature which
constitutes the background of the present work, highlighting similarities and divergences, and stressing the
potential benefits arising from the combination of the Agent Based and Stock Flow Consistent approaches.
In section 3 we sketch out the basic structure of the model, the sequence of events taking place within each
round of the simulation, and the features of the matching mechanism adopted on our simulated markets.
Section 4.1 presents the behavioral equations of consumption and capital good producers. Section 4.2
instead presents the heuristics followed by banks, with a particular focus on their credit supply function.
The causes and consequences of firms and banks failures are treated in section 4.3, while sections 4.4
and 4.5 present the behaviors of households, the Government, and the Central Bank. In section 5 we
explain the logic adopted to setup the simulations while section 6 discusses the results of our simulations
and the properties of our simulated economy. In the conclusion we discuss the open issues of our research
and briefly schedule future works.

Related literature

Several authors have advocated increasing investment in agent-based modeling in response to the crisis
(Farmer and Foley, 2009), (Colander et al., 2009). The agent-based approach is rooted in the conception
of the economy as a complex adaptive system composed of heterogeneous, boundedly rational agents,
interacting locally in a given institutional framework (see Esptein (2006) for a general discussion of the
methodology). Contributions in this field highlight how even the simplest microeconomic behaviors may
lead to complex systemic properties due to feedbacks, externalities, and other structural effects arising
from agents disperse interactions. For example, Brock and Hommes (1997) focuses on the heterogeneity
of agents beliefs and selection mechanisms of agents heuristics to explain the convergence towards wellbehaved equilibria, or the emergence of chaotic dynamics in financial markets. Along the same path,
Chiarella and He (2001) incorporated in an equilibrium model the interactions of heterogeneous agents
in financial markets, showing how the resulting dynamical system for asset price and wealth turned out
to be non-stationary.
Empirically, agent based macroeconomic models are capable of reproducing a significant number of
micro and macroeconomic stylized facts (see, for example, Dosi et al., 2010, 2013, 2015; Delli Gatti
et al., 2008; Assenza et al., 2015; Riccetti et al., 2014), often outperforming DSGE models (Fagiolo
and Roventini, 2012). In particular, Agent Based models have proven to be well-suited to explain the
emergence of financial fragility. Much attention has been devoted to the analysis the impact exerted on
business cycles by credit conditions and firms finance, in a context of incomplete asymmetric information
and imperfect financial markets (Greenwald and Stiglitz, 1993). Good examples of this field of research
are Delli Gatti et al. (2005, 2008, 2010b) who focused on the role of commercial and banks credit networks
topology in spreading financial fragility through contagion effects.
Similarly, Cincotti et al. (2010) investigates the link between business cycles and monetary aggregates,
finding that the amplitude of business cycles is greater the more firms resort to external finance, rather
than internal funding. Using the same model, Raberto et al. (2012) show that debt dynamics plays a key
role in shaping business cycles dynamics: exogenous changes to the regulatory capital requirement foster
higher indebtedness which generates boom and burst dynamics.
Our attention is not limited to the agent-based literature. We point to the so called flows-of-funds
approach Godley and Cripps (1983) which aims at providing a comprehensive and fully integrated representation of the economy, including all financial transactions. Using flow of funds accounts to analyze the
US economy at the turn of the century, Godley and Wray (1999); Godley and Zezza (2006) pointed out
that growing households indebtedness was pushing assets inflation and leavening systemic risk under the
surface of the alleged stability of the early 00s, thereby anticipating the crisis with significant precision
regarding the timing and mechanics of the collapse.
In 2011, the Bank of England used a flow-of-funds approach to analyze the mechanics of financial
4

instability. Barwell and Burrows (2011) advocated the diffusion of macroeconomic approaches that stress
the importance of balance sheet linkages in spotting buildups of financial fragility. Stock Flow Consistent
(SFC hereafter) models, stemming from Godleys earlier work, aim at responding to this call (Godley,
1997; Godley and Lavoie, 2007). The objective of the SFC framework is to provide a comprehensive and
fully integrated representation of the real and financial sides of the economy through the adoption of
rigorous accounting rules based on the quadruple entry principle developed by Copeland (1949). This
approach employs specific social accounting matrices to track the variation of financial stocks and flows,
and to ensure that every financial stocks or flow in the economy is recorded as a liability or outflow for
someone and an asset or inflow for someone.2
Our feeling is that AB models may greatly benefit from an integration with the SFC accounting
framework. As noted by (Bezemer, 2012, p.18)
. . . complex behaviors and sudden transitions also arise from the economys financial structure
as reflected in its balance sheets.
We feel that a fusion of the two approaches has the potential to set an alternative paradigm to economic
modeling, as advocated by Farmer and Foley (2009); Delli Gatti et al. (2010a).
AB models can help to overcome many of the limitations of the SFC approach (Kinsella, 2011). For
example, SFC models are traditionally highly aggregated, dividing the economy in major institutional
sectors, typically households, banks, firms, and the public sector.3 This perspective usually abstracts
from tracking intra-sectoral flows and does not allow to analyze the causes and effects of agents heterogeneity emerging within and across sectors. This limit definitively hinders, and in some cases impedes,
the possibility of studying phenomena which are deeply connected to agents heterogeneity and agents
disperse interaction, such as self-organization processes within markets or industries, and the generation
of financial bubbles,
To our knowledge, the economic literature provides just one other example of AB model implementing
SFC starting from the very bottom layer, represented by individual agents balance sheets: the EURACE
model (see Deissenberg et al., 2008; Cincotti et al., 2010; Raberto et al., 2012; Dawid et al., 2012, 2014;
van der Hoog and Dawid, 2015), a massively large-scale economic model of the EU economy first developed in 2006 and now implementing many hyper-realistic features such as day-by-day interactions,
asynchronous decisions, different frequencies of economic processes, geographical space, and a huge variety of agents, including even international statistical offices. Although our work shows several points
of convergence with the philosophy of the EURACE project, our objective is different and somehow
complementary.
Our work started from the consideration that, although a few attempts to develop AB-SFC models
exist in the literature (Kinsella et al., 2011; Riccetti et al., 2014; Seppecher, 2012), we have not yet a
well-defined set of concepts, rules and tools to develop these models. A similar criticism may partially
apply also the EURACE model, as its main strengths also tend to limit its accessibility and re-usability
by other scholars. The number of complex features implemented in the EURACE model may make it
difficult for those who are not directly involved in its development to enter the logic of the model, let alone
master the skill required to manage it, and adapt it for new purposes. Furthermore, running such a large
model necessitates the use of massively parallel computing clusters, which are not generally available to
all scholars.
Our objective differs in that we do not want to move towards a one-to-one matching of real economies,
but rather we aim to develop a relatively simple and flexible AB-SFC model to study policies and
institutional setups to steer economic cycles and tackle economic instability. Our contribution aspires to
serve as a benchmark for all scholars in the AB and SFC communities, as well as from other modeling
traditions, who want to get involved in this approach.4 .
2 By looking at the evolution of aggregate inter-sectoral flows and stock-flow norms, one can then get a first clue of the
possible emergence of imbalances in the stocks accumulation between sectors, eventually leading to crisis if not properly
tackled.
3 See Caverzasi and Godin (2015) for a recent survey of the SFC literature.
4 Our model has been developed using our brand new Java Macro Agent Based (JMAB) programming tool suite, explicitly
designed for AB-SFC models. The platform exploits the logic of the object-oriented programming paradigm to provide a
flexible and highly modular computational framework embedding general procedures to ensure and check the model stockflow consistency at the macro, meso and micro levels. It has the potential to implement a wide variety of models and a
number of crucial features of modern economic systems, in particular regarding the handling of heterogeneous real and
financial stocks in agents balance sheets (see for example section 4.2) and the representation of commercial and financial
networks.

The model

Figure 1: Flow Diagram of the model. Arrows point from paying sectors to receiving sectors.

The economy described by the flow diagram of figure 1 is composed of various groups, or sectors,
of boundedly rational agents, following simple heuristics in a context of incomplete and asymmetric
information. More precisely, the model contains:
A collection H of households selling their labor to firms in exchange for wages, consuming and
saving in the form of banks deposits. Households own firms and banks proportionally to their
wealth, and receive a share of firms and banks profits as dividends. Unemployed workers receive
a dole from the government. Finally, households pay taxes on their gross income.
Two collections of firms: consumption (C ) and capital (K ) firms. Consumption firms produce
a homogeneous consumption good using labor and capital goods manufactured by capital firms.
Capital firms produce a homogeneous capital good characterized by the binary {k , lk }, indicating
respectively the capital productivity and the capital-labor ratio. Firms may apply for loans to
banks in order to finance production and investment. Retained profits are held in the form of
banks deposits.
A collection B of banks, collecting deposits from households and firms, granting loans to firms,
and buying bonds issued by the Government. Mandatory capital and liquidity ratios constraints
apply. Banks may ask for cash advances to the Central Bank in order to restore the mandatory
liquidity ratio.
A Government sector, which hires public workers (a constant share of the workforce) and pay
unemployment benefits to households. The government collects taxes and issues bonds to cover its
deficits.

 A Central Bank, which issues legal currency, holds banks reserve accounts and the government
account, accommodates banks demand for cash advances at a fixed discount rate, and possibly buy
government bonds which have not been purchased by banks.

3.1

Dispersed interactions

During each period of the simulation agents interact on five markets:

A consumption goods market: households interact with consumption firms;
A capital goods market: consumption firms interacts with capital firms;
A labor market: households interact with government and both types of firms;
A credit market: firms interact with banks;
A deposit market: households and firms interact with banks.
Following Riccetti et al. (2014), we explicitly model agents dispersed interactions by assuming that
agents on the demand and supply sides of each market interact through a common matching protocol.
In each period of the simulation, demand agents are allowed to observe the prices or the interest rates
charged by a random subset (whose size depends on a parameter proxying the degree of imperfect
information) of potential suppliers. They choose the best one: the cheapest counterpart for goods,
labor, or credit markets, or the bank offering the highest interest rate for the deposit market. Each agent
then has a probability of switching (P rs ) from the previous supplier to the new one defined as follows
(Delli Gatti et al., 2010a). For the consumption, capital, and credit markets, where prices (or interest
rates) express a disbursement from the demander, the probability of switching to the new partner is
decreasing (in a non-linear way) with the difference between pold and pnew :

1 e(pnew pold )/(pnew ) if pnew < pold
P rs =
(3.1)
0
otherwise
On the deposit market, interest rates generates an income for the depositor, the probability of switching
is thus:

1 e(pold pnew )/(pold ) if pnew > pold
P rs =
(3.2)
0
otherwise,
where > 0 is an exogenous parameter.
In some cases, some suppliers exhaust inventories available for sale, possibly leaving some customers
with a positive residual demand. We then allow demand agents to look for other suppliers within the
original random subset of potential partners in order to fulfil it. Markets interactions are closed when
demand agents have fulfilled their demand, when there are no supply agents willing or able to satisfy
their demand, or if demanders run out of deposits to pay for demanded goods.

3.2

Sequence of events

In each period of the simulation, the following sequence of events takes place:
1. Production planning: consumption and capital firms compute their desired output level based on
their sales expectations (sec: 4.1.1).
2. Firms labor demand : firms evaluate the number of workers needed to produce the desired level of
output (sec: 4.1.1).
3. Prices, interest, and Wages: consumption and capital firms set the price of their output while
banks determine the interest rate on loans and deposits (sec: 4.1.2 and 4.2). Workers adaptively
revise their reservation wages (sec: 4.4).
4. Investment in capital accumulation: consumption firms determine their desired rate of capacity
growth and, as a consequence, their real demand for capital goods
5. Capital good market - first interaction: consumption firms interact with capital firms and choose
their prefered supplier.
7

6. Credit demand : firms assess their demand for credit based on their expected financial requirement
and their available internal funds.
7. Credit market: interaction takes place on the credit market. Banks evaluate each loan request and
decide whether to grant the whole amount, a share of it, or nothing. The loan creation process
leads to an expansion of the banks balance sheets, as the new loan recorded as an asset is mirrored
by a new deposit, a liability for the bank.
8. Labor market interactions: unemployed workers interact with firms on the labor market (sec: 4.5).
9. Production: capital and consumption firms produce their output.
10. Capital goods market - second interaction: consumption firms purchase capital goods from their
prefered supplier. Newly purchased machineries can be employed in the production process starting
from the next period.
11. Consumption goods market: households decide how much to consume depending on their expected
real disposable income and expected real wealth, with fixed propensities (sec: 4.4).
12. Interest, bonds and loans repayment: firms pay interest on loans and repay a (constant) share of
each loan principal (sec: 4.1.6). The government repays bonds and interest to bonds holders. Banks
pay interest on deposits. If in the previous period they asked for cash advances to the Central Bank,
they repay this amount plus a fixed exogenous interest.
13. Wages and dole: wages are paid to employed workers. Workers who are not employed receive a
dole from the government.
14. Taxes: taxes on profits (of firms and banks) and income (of households) are paid to the government
(sec: 4.1.6).
15. Dividends: a constant share of net-profits realized by banks and firms are distributed to households
proportionally to their net wealth.
16. Deposit market interaction: households an firms decide in which banks to deposit their savings
17. Bond purchases: the government issues new bonds to cover any deficit between expenditure and
revenues (sec: 4.5). We assume the interest rate on bonds and their price to be fixed. Banks use
part of their reserves to buy bonds. The Central Bank buys all residual bonds.
18. Cash Advances: banks may ask the Central Bank for cash advances in order to respect the mandatory reserve ratio.
In each period of the simulation, firms may default when they run out of liquidity to pay wages or to
honor the debt service while banks default if their net wealth turns negative. The effects of firms and
banks defaults are treated in section 4.3.

Agent behaviors

This section details the behavior of each type of agents. We used the following notation in the equations.
Consumption firms variables have a c subscript, capital firms a k, households a h and banks a b. If the
variable is identical for consumption and capital firms, we used the x subscript. All agents share the same
simple adaptive scheme to compute expectations (indicated by a e superscript) for a generic variable z:
e
e
zte = zt1
+ (zt1 zt1
)

4.1
4.1.1

(4.1)

Firm behavior
Production planning and labor demand

D
Firm x desired output in period t (yxt
) depends on the firms sales expectations sext . We assume firms want
to hold a certain amount of real inventories, expressed as a share of expected sales, as a buffer against

unexpected demand swings (Steindl, 1952) and to avoid frustrating customers with supply constraints
(Lavoie, 1992).
D
yxt
= sext (1 + ) invxt1 with x = {c, k}

(4.2)

Firms in the capital-good industry produce their output out of labor only. Capital firms demand for
D
workers depends on ykt
and the labor productivity N , which we assume to be constant and exogenous.
D
D
Nkt
= ykt
/N

(4.3)

The labor requirement of any consumption firm c can be calculated as:

D
Nct
= uD
ct

kct
.
lk

(4.4)

D
where uD
ct is the rate of capacity utilization needed to produce the desired level of output yct :

uD
ct = M in(1,

D
yct
), ,
kct K

(4.5)

where kct indicates the real stock of capital. The technology embedded in capital goods is defined by the
binary {K , lk } where the former parameter indicates capital productivity and the latter the constant
capital-labor ratio.
We assume a positive employee turnover, expressed as a share of firms employees (sec.4.1.6). Firms
then determine wether they need to hire new workers or fire some of them. Redundant workers are
randomly sampled from the pool of firm employees.
4.1.2

Pricing

Prices of goods are set as a non-negative markup muxt over expected unit labor costs:
pxt = (1 + muxt )

e
D
Wxt
Nxt
,
D
yxt

(4.6)

e
is the expected average wage.
where Wxt
The mark up is endogenously revised from period to period following a simple adaptive rule. When
firms end up having more inventories than desired (see sec.4.1.1), the markup is lowered in the next
period, in order to increase the attractiveness of their output.
(
xt1
muxt1 (1 + F N ) if inv
sit1
muxt =
(4.7)
xt1
muxt1 (1 F N ) if inv
sit1 > ,

where F N is a random number picked from a Folded Normal distribution with parameters (F N , F2 N ).
4.1.3

Firms profits

Firms pre-tax profits are the sum of revenues from sales, interest received, and the nominal variation
of inventories,5 minus wages, interest paid on the collection of outstanding loans, and eventually capital
amortization flows.
Formally, for consumption firms:
ct = sct pct + idbt1 Dct1 + (invct ucct invct1 ucct1 ) . . .
...

nNct

wnt

t1
X

j=t

ilj Lcj

[(t 1) j]

kKct1

(k k pk )

(4.8)

where idbt1 is the interest rate on past period deposit Dct1 held at bank b, ucc are unit costs of
production, wnt is the wage paid to worker n, ilj is the interest rate on loan Lcj obtained in period
5 In accordance with standard accounting rules, firms inventories are evaluated at the firms current unit cost of production. As a consequence, the value of inventories may vary due to variation of either their quantity or of their productive
costs.

j = t , ..., t 1, pk is the price paid for the batch of capital goods k k belonging to the firms collection
of capital goods Kct1 , and = are the duration of loans and capital respectively. Similarly, capital
firms profits are given by:
kt = skt pkt + idbt1 Dkt1 + (invkt uckt invkt1 uckt1 ) . . .
...

wnt

t1
X

j=t

nNkt

ilj Lkj

[(t 1) j]

(4.9)

This accounting definition of profits is then used to compute the amount of taxes firms have to pay:
Txt = M ax { xt , 0}, being the corporate profits tax rate. We also consider an alternative measure
of firms performance in order to capture the actual ability of the firm to generate cash inflows (which
can be accumulated or used to finance investment or to distribute dividends) through its normal business
operation.6 We define the operating cash flow as their income out of sales or interest perceived minus
the wage bill and total debt service (i.e. interest and principal payments):
OCFxt = sxt pxt + idbt1 Dxt1

nNxt

wnt

t1 
X

j=t

ilj Lxj

[(t 1) j] Lxj
+

(4.10)

Notice that the operating cash flow can be interpreted as a sort of Minskian litmus paper to assess
whether a firm is in a hedge, speculative, or Ponzi financial position. An OCF 0 implies that the
firm is capable of enough generating cash flow to honor the debt service (hedge position). If the OCF is
negative, but its absolute value is less than or equal to the principal repayment, the firm is in a speculative
position since its cash flows are sufficient to cover the interest due, but the firm must roll over part or all
of its debt. Finally, when the OCF is negative and its absolute value is greater than principal payments,
the firm is trapped in a Ponzi position.
4.1.4

Investment

D
Firms invest in each period in order to attain a desired productive capacity rate of growth gct
which
D
depends on two ratios: the desired rate of capacity utilization uct (4.5) and the past period value of a
modified rate of profit rct , based on operating cash flows rather than on accounting profits.

rct1 r
uD u
+ 2 ct
r
u
OCFct
rct = P
agekt1 .
k k
)
kKct1 (k p )(1

D
gct
= 1

(4.11)
(4.12)

Here, u and r denote firms normal rates of capacity utilization and profit respectively, both assumed
to be constant and equal across firms. The denominator in equation (4.12) expresses the previous period
value of the firms stock of capital, with agekt1 indicating the age in period t 1 of the batch of capital
goods k belonging to the collection Kct of firm c.
D
, we can derive the real demand for capital goods iD
Given gct
ct as the number of capital units required
to replace the obsolete capital7 , and to fill the gap between current and desired productive capacity level.
D
Once firms have chosen their capital good suppliers, nominal desired investment Ict
can be computed by
D
multiplying ict for the price pkt applied by the selected supplier k.
4.1.5

Firms finance

Since Fazzari et al. (1988), more and more empirical evidence contradicting the Modigliani and Miller
(1958) theorem which predicts that, at the margin, alternative sources of finance should be perfect
substitutes have been gathered. Solid arguments have been provided in favor of a pecking order theory
of finance (Meyers, 1984), according to which firms mainly rely on their retain earnings and only resort
to external financing when internal funding possibilities have been completely exhausted. In the presence
6 The

accounting definition of profits includes accounting items, such as the variation of nominal inventories and capital
amortization, which are not matched by any actual cash flow during the period, while it omits others which do give rise to
cash flows, such as principal repayments of outstanding loans.
7 For sake of realism, we assume that the financial value of each capital batch is lowered by a constant share (1/) of the
original purchasing value in each period, while the correspondent real stock of machinery can be used at full potential till
it reaches age =

10

of imperfections in capital markets (e.g. information asymmetries), the cost of external finance (equity
emissions and loans) is usually high. Credit demand should then be defined as the difference between
expected net disbursement and available internal funds. However, firms almost never arrive to the point
of exhausting all their internal resources before asking for credit. We therefore adopt a slightly modified
version of the pecking order theory adopted in most macro-AB models, by assuming that firms desire
to hold a certain amount of deposits, expressed as a share of the expected wages disbursement, for
precautionary reasons. The demand for credit by consumption firms is then:
D
e
e
D
e
LD
ct = Ict + Divct + Wct Nct OCFct

(4.13)

e
where Divct
are expected dividends distributed to equity holders (i.e. households). Actual dividends are
computed as a constant share x of firms expected after-tax profits: Divxt = M ax {0, x xt (1 )}.
The credit demand function for capital firms can be easily derived from equation (4.13) by simply omitting
ID.

4.1.6

Labor, Goods and Deposit markets

After the credit market interaction between banks and firms has taken place, firms interact with unemployed households on the labor market. Production then takes place. Firms output can be constrained
by the scarcity of available workers (i.e. full employment case) or, in the case of consumption firms, by
their stock of capital (i.e. capacity constraints). Once the production process has been accomplished
households and consumption firms interact on the consumption good market.
The second phase of the market interaction on capital market then takes place with consumption
firms buying capital goods previously ordered (see sec.4.1.4).
Finally, gross profits xt are computed, taxes Txt are paid, and dividends Divxt distributed to households (see sec.4.4).

4.2

Bank behavior

In our model, every time new credit is granted, two accounting items are added to the balance sheets of the
bank and the borrowing firm: a loan and a correspondent deposit. Despite the story still described in many
economic textbooks, today it is widely recognized (see for example the two authoritative contributions
by Benes et al. (2014) and McLeay et al. (2014)) that in the process of making new loans, commercial
banks create matching liabilities in the form of bank deposits for their borrowers, thereby expanding their
balance sheets. The main implication of the credit creation process is that bank loans give borrowers
new purchasing power, that is create new means of payment, that did not previously exist.
Another crucial aspect of the credit creation process, which is often dismissed, is that loans represent a
long term commitment both for borrowers and lenders. Banks full flexibility to expand their balance sheet
while they are constrained when they want to downsize it as loan contracts contain detailed commitments
regarding amortization schedules. Most of the risk lenders have to bear originates from this temporal
dimension of credit. On the other hand, it should be straightforward that banks ability to create money
through new loans is limited by the need to remain profitable in a competitive banking system.
Long term duration of loans and a financial system composed of multiple competing banks should then
be two fundamental features of every model aiming at studying the role of credit in modern macroeconomic
systems.
Nonetheless, the vast majority of previous macroeconomic computational models either assume loans
to have a one period maturity, or they assume a single giant bank, or that the credit network of firmsbanks is pre-determined once and for all in the simulation set-up.8 In our model, we assume loans to last
for = 20 periods (i.e. 5 years)9 and we allow firms to interact with several banks on the credit market
during each simulation round.10 From period to period, firms may switch from one bank to another so
that they will generally have a collection of heterogeneous loans with different banks on the liability side
of their balance sheets.
8 We believe that the main reason for these simplifying assumptions is not to be found on the theoretical level, but rather
in the technical difficulties to handle the multiplicity of heterogeneous loans characterized by different age, different interest
rates, different liability and asset holders, etc. JMAB effectively overcomes these difficulties by exploiting the opportunities
of Object Oriented Programming.
9 Loans are repaid following the same amortization scheme: in each period firms repay a constant share (1/) of the
original amount.
10 To our knowledge, the only model sharing these features is the Eurace/Eurace U niBi model (Raberto et al., 2012;
van der Hoog and Dawid, 2015)

11

On the supply side, banks can grant several loans to different firms within the same period. Our
matching mechanism on the credit market therefore implies that lenders and borrowers are linked through
a complex bipartite network endogenously evolving through the simulation.
Another crucial aspect of our modeling approach to credit market regards the implementation of
the credit rationing mechanism. In the earliest AB macroeconomic literature, banks were assumed
to accommodate loan requests by borrowers, eventually discriminating borrowers only through interest
rates.11 However, in reality banks mainly discriminate through credit rationing rather than interest rates.
Recently, and clearly in response to the crisis, the AB literature has started to introduce some credit
rationing mechanisms. A simple rationing mechanism is presented in Riccetti et al. (2014) which assume
banks want to avoid being too exposed towards single borrowers, thereby setting an upper bound to loans
that can be grant to single borrowers, expressed as a share of total loans. Dosi et al. (2013) assume a
unique bank willing to grant loans up to a given multiple of deposits in each period. When loan requests
by firms exceed this maximum level of available credit, those who are badly ranked by banks will be credit
constrained. A similar ranking mechanism is also presented in van der Hoog and Dawid (2015), although
they have multiple banks willing to accommodates loan requests as long as their overall outstanding
credit is compatible with the capital ratio institutional constraint. This feature is also shared by Raberto
et al. (2012) where banks accommodates loans as long as the risk-weighted credit demanded by firms
is compatible with capital adequacy ratios. Assenza et al. (2015) instead assume there is a maximum
admissible loss on each loan, expressed as a share of banks equity. Given the borrowers estimated
probability of default, they define an upper bound to the amount of funds the bank is willing to lend.
In all these credit-rationing mechanisms, the opportunity of a loan tend to be evaluated by relating
the riskiness added to the banks portfolio of assets by the new loan to the banks current overall financial
position. While this somehow reflects the fact that single loan requests must be evaluated also in relation
to the banks overall balance sheet management, we believe that a proper credit-rationing mechanism
should also take into consideration the expected internal rate of return associated to each specific credit
application.
Most of the computational economics literature presenting explicit credit riskiness evaluation mechanisms still uses only stock indexes to assess borrower reliability. In many cases, the leverage ratio is
the unique measure used to compute the probability to default of credit applicants. This, we feel, is not
enough do understand the mechanisms truly at play. We do not deny that high debt-to-equity ratios may
rise an alarm bell, reflecting a deterioration of borrowers financial soundness. It must be pointed out
however that the leverage ratio, per se, does not provide any information about borrowers actual ability
to generate net cash inflows to honor the debt. We therefore present a different procedure to determine
credit supply of the banks, based on the following three pillars:
Active management of banks balance sheet: in keeping with much of the literature, we assume
banks credit supply must be somehow related to the overall bank financial position. In our model
each bank has an endogenously evolving capital ratio target, which for simplicity reasons is assumed
to be equal across banks and defined as the average of the sector. Furthermore, we also assume
banks capital ratio has a mandatory lower bound (6%). Banks thus manage their balance sheets in
order to pursue their targeted capital ratio through the interest rate strategy, affecting the number
of potential customers.
Case-by-case credit worthiness evaluation: banks willingness to satisfy fully, partially, or not at all
individual companies credit demand is based on a comparison between expected cash inflows and
expected losses, given the applicants probability of default and its collateral value.
Credit worthiness based on operating cash flows. Banks compute borrowers uni-periodal probability
of default, under the hypothesis a loan is granted, by comparing firms current operating cash flows
with the single period payments to serve the debt and repay a share of the principal.
Banks interest rates on loans depends on the comparison between banks current capital ratio CRbt =
N Wbt /LTbtot and the common target CRtT . When the capital ratio is higher than its target, banks are
more capitalized than needed and may expand further their balance sheet by granting more loans.Their
ability to attract customers is increased by offering a lower interest rate lower than their competitors.
11 In the standard framework, loan interest rate were defined as the sum of a bank-specific component, usually related
to the banks financial position (proxied by its net-worth), or an institutionally set base loan rate, and a borrower-specific
component (risk premium) increasing with some measures of the borrowers riskiness, typically some net-worth related
indexes such as the leverage ratio. See for example Delli Gatti et al. (2010b); Raberto et al. (2012).

12

In the opposite case, a higher interest rate has the twofold effect of making banks loans less attractive,
while increasing the banks margin. Formally:
( l
ibt1 (1 + F N ) if CRbt < CRtT
l
(4.14)
ibt =
l
ibt1 (1 F N ) otherwise ,
where

l
ibt1

ilbt1
sizeB

bB

is the market average interest rate in the previous period and F N is a realization

of a stochastic variable distributed according to a Folded Normal Distribution (F N , F2 N ).

Case-by-case credit rationing mechanism starts with banks evaluating applicants single-period probability of default, under the hypothesis thats the loan requested is granted. We define the debt service
d
variable as the first tranche of payment associated to the hypothetic loan: dsL = (ilbt + 1 )Ld . The probability of default is then computed using a logistic function, based on the percentage difference between
d
borrowers OCFxt and dsL :
prxD =

1
1+

Ld
x ds
exp( OCFxt
dsLd

(4.15)

c and k are two parameters expressing banks risk aversion in lending to capital and consumption firms.
The higher the more banks are risk averse (i.e. the higher the probability of default for given OCF and
ds).
The use of the OCF to assess borrowers reliability has the advantage, highlighted in section 4.1.3,
of providing a clear measure of the borrower financial position if the loan was granted. As the OCF
accounts both for interest and principal payments on the existing stock of debt, they allow to capture
the effect of high indebtedness just as the leverage ratio, but also providing a criterion to rank potential
borrowers having the same leverage but different performances.
The expected return of a credit project also depends on firms collateral, when present. Our model
assumes that consumption firms can use their stock of real capital as collateral. In the case of default,
each bank expects to be able to recover a share x of their outstanding loans to the defaulted firm x
through fire sales of its capital.12 Knowing Ld , ilbt , prxD , x , we can compute the overall expected return
of a credit project by simply summing up the payoffs arising from each possible outcome of the decision
to grant the loan, each one weighted for its probability of occurrence. Figure 4.2 provides a graphical
representation of the payoffs tree.
LOAN GRANTED

prD

Credit Payoffs:

(1-prD)

=4
=1/
=K Discounted Value/Total Debt
-Ld
DEFAULT

prD

-[(Ld- Ld)(1- )-iLd]

DEFAULT

(1-prD)2

prD

(1-prD)3

-[(Ld-2 Ld) (1- )-iLd-i(1- )Ld]

prD

DEFAULT

-[(Ld-3 Ld)(1- )-iLd-i(1- )Ld-i(1-2 )Ld)]

DEFAULT

(1-prD)4

(iL+i(1- )L+i(1-2 )L+i(1-3 )L)

LOAN REPAID

12 Notice that, since capital good producers employ only labor in their production process and do not invest in capital
accumulation, they have no collateral.

13

Banks are willing to satisfy agents demand for credit whenever the expected return is greater or equal
than zero. Otherwise, the bank may still be willing to provide some credit, if there exist an amount LD
for which the expected return is non-negative.
Deposits and bonds market
Banks hold deposits of households and firms. Individual banks amount of deposits depends on two factors:
first, firms and households decisions about where to put their savings; second, agents autonomous
transactions realized through deposit transfers. In this respect, it must be noticed that deposit transfers
between different banks also implies a clearing mechanism of transfer of reserves from the paying to the
receiving bank.
As banks have to satisfy mandatory liquidity ratios (8%) and since deposits represent a source of
a
reserves much cheaper than Central Bank cash advances (that is, idbt << icb ) banks compete with each
other on the deposit market.13 As in the case of the capital ratio, we assume that banks have, besides the
mandatory lower bound, a common liquidity target LRtT defined as the sector average in the last period.
In order to attain this target ratio banks manage the interest rate offered on deposits according to a
simple heuristic similar to that presented in equation 4.14. When the liquidity ratio is below the target
banks set their interest on deposits as stochastic premium over the average interest rate in order to attract
customers, and vice-versa when banks have plenty of liquidity.
( d
ibt1 (1 F N ) if LRbt LRtT
d
ibt =
(4.16)
d
ibt1 (1 + F N ) otherwise ,
where F N indicates the realization of a stochastic variable drawn from a Folded Normal Distribution
(F N , F2 N ).
Finally, we assume that banks use their reserves in excess of their target to buy government bonds.14
All bonds not purchased by commercial banks are bought by the Central Bank.

4.3

Firms and banks bankruptcy

Firms and banks may go bankrupt when they run out of liquidity or if their net-wealth turns negative.
For simplicity reasons, we assume defaulted firms and banks to be bailed in by households (who are the
owners of firms and banks and receive dividends) and depositors in order to maintain the number of firms
and banks constant.
A bankruptcy by a firm induces non-performing loans for its creditors, who see their net wealth
shrinking. In the case of capital firms, the loss is totally borne by banks, as capital firms do not have
any collateral. In the case of a consumption firm, we assume that its ownership passes temporarily to
creditors which try to recover part of their outstanding loans through fire sales of firms physical capital
to households.
In accordance with empirical evidence, we assume the financial value of assets sold through fire sales
to be lowered by a share , which is assumed to be an exogenous parameter. When the discounted value
of capital is greater or equal to the firms bad debt, the loss caused by the bankruptcy falls completely
on households shoulders while banks are able to recover all their loans.
However, in the general case, the loss is split between households and banks which are able to recover
only a fraction of their loans. Individual households contribution to fire sales follows the same rule
of dividends distribution (sec.4.4), with the overall disbursement being distributed proportionally to
households net wealth.
The revenues from sales of the firms assets are then distributed across creditors proportionally to
the fraction of firms total debt they own. It is trivial to show that, given these assumptions, creditors
are able to recover the same share c < 1 of their loans, expressed by the ratio between firms capital
13 Whenever the capital ratio falls below the mandatory threshold banks apply for cash advances to the Central Bank
(see 4.5).
14 Remember that, given the sequence of events in each simulation round, government bonds repayment occurs after
deposit interest rates have been set and deposit market interactions have taken place. This bonds repayments increase
banks reserves providing extra liquidity that banks use to purchase newly issued bonds. In the unlikely case banks are
willing to buy more bonds than available, these are distributed across banks proportionally to their demand (i.e. to the
employable amount of reserves).

14

discounted financial value and total bad debt:

P
[ kKct (k k pk )(1 agekt )]
c = Pt
.
[(t1)j]
l
j=t ij Lcj

(4.17)

The event of the bankruptcy by a bank, though unlikely, is treated here by assuming that depositors
bear the loss associated to the default. In order to restore a positive net-wealth, deposits are lowered up
to the point the banks capital ratio equals the minimum capital adequacy requirement (6%), similar to
a bail-in process. The total loss borne by the depositor is distributed proportionally to the scale of their
deposits.

4.4

Household behavior

Workers follow an adaptive heuristic to set the wage they ask for: if over the year (i.e., four periods), they
have been unemployed for more than two quarters, they lower the asked wage by a stochastic amount.
In the opposite case, they increase their asked wage, provided that the aggregate rate of unemployment
in the previous period (ut1 ) is sufficiently low. This latter condition is meant to mimic the endogenous
evolution of workers bargaining power in relation to employment dynamics.
 D
P4
wht1 (1 F N ) if
uhtn > 2
Pn=1
wid,t =
(4.18)
4
D
wht1 (1 + F N ) if
n=1 uhtn 2 and ut1 ,

where uht = 1 if h is unemployed in t, and 0 otherwise.

Workers decide how much to consume with fixed propensities 1 , 2 out of expected real disposable
income and expected real wealth. As workers set their real demand before interacting with consumption
firms, they construct expectation on consumption good prices peht :
cD
ht = 1

N Wht
N Iht
+ 2 e
e
pht
pht

(4.19)

Gross nominal income is given by wht + idbt1 Dht1 + Divht if the worker is employed. Households
pay taxes on income with a flat tax rate i . Unemployed workers receive a tax-exempt dole from the
government defined as a share of average wages.

4.5

Government and central bank behavior

The government hires a constant share of households. Public servants are also subject to a turnover .
Furthermore the government pays unemployment benefits (dt ) to unemployed people (Ut ), acting in a
counter-cyclical way.
The state collects taxes on income and profits (with constant rates i and ) from households, firms
b
and banks and issues bonds bt (at fixed price pb and interest i ) to cover the eventual deficit between
expenses and revenues. Bonds are assumed to last 1 period for simplicity reasons:
X
b
wn Ut dt i pb bt1 ,
pb bt = Tt + CBt
(4.20)
nNgt

where Tt = THt + TCt + TKt + THt are total taxes, CBt are Central Bank profits, Ngt is the collection
of public workers.
Bonds not purchased by commercial banks are bought by the Central Bank. Notice that Central
Bank purchases of government bonds are one of the two channel by which reserves are injected into
banks balance sheets, the other one being represented by cash advances which the Central Bank grants
on request by commercial banks. Cash advances are assumed to be repaid after one period and their
constant interest rate represents the upper bound for interest paid by banks on customers deposits. For
simplicity reasons, we assume the Central Bank pay no interest on banks reserves account. Finally,
Central Bank earns a profit equal to the flow of interest on its holdings of government bonds and cash
advances:
b

CBt = i Bt1 + iCB CAcbt .

As mentioned above, Central Banks profits are distributed to the government.
15

(4.21)

Baseline setup: challenges in calibration

Calibration represents a crucial issue for every computational model, in particular whenever they entail
stochastic, path-dependent, non-ergodic dynamics. Monte Carlo and sensitivity analysis techniques are
an essential element in order to assess across-runs variability, to perform robustness checks and policy
analysis, and to validate simulation models. This is particularly true for theoretical micro-founded models
for which micro-data are often lacking, or non-observable at all.
Before describing our approach to setup the model baseline, we highlight three main challenges related
to the calibration process: (i) setting a clear and replicable procedure trying to minimize the arbitrariness
in defining parameters values. (ii) making sure that the economy as a whole starts from stock-flow
consistent initial conditions iii) define a strategy to characterize and distribute agent-specific stocks
starting from aggregate ones.
Technical difficulties, time, and computational limits often prevent the modeller from exploring the
entire parameter space, in particular for large-complex macroeconomic model. These models are thus
usually analyzed in the local neighborhood of a baseline scenario. Setting up an adequate baseline implies
the need to find a strategy to constrain the parameter space, to limit the number of free parameters,
and to distribute initial endowments across agents in a way such that initial conditions do not entail any
a-priori bias for the phenomena we want to analyze through the simulation experiments. Still, only a few
articles presenting computational macroeconomic models provide an exhaustive explanation of the logic
followed to calibrate their baseline.
As highlighted in the previous sections, one of the main aspects of novelty of the present work is
the implementation of stock flow consistent accounting rules in the bottom-up perspective of agent based
models. It is thus important to start from initial conditions respecting a stock-flow consistent distribution
of aggregate real and financial stocks across sectors. That is, we shall find a procedure to determine jointly
the initial values of different types of stocks held by each sector, respecting Copelands quadruple entry
principle.
Aggregates stocks should then be distributed across agents within each specific sector of the economy,
thus specifying the overall shape of agents initial balance sheets. As described in the previous sections,
agents balance sheets are sometimes characterized by the presence of multiple stocks of the same types,
which differs in terms of quantity, age, maturity, prices, and liability & asset counterparts. In our
model, this is the case for loans in firms and banks balance sheets, and capital goods in consumption
firms balance sheets (see sec.4.2 and 4.1.4). The third challenge thus consists thus in finding a coherent
strategy to characterize each stock in these collections and assign them to each individual agent who hold
them as an asset or a liability.
In order to respond to the three challenges, we adopted the following six-step strategy:
1. We derive an aggregate version of the model.
2. We constrain the aggregate model15 to be in a real stationary state associated with a nominal
steady growth equal to gss . This imply that while all real quantities are constant, all prices and
wages are growing at the same rate gss .16
3. We numerically solve the constrained model by setting exogenously reasonable values for the parameters for which some empirical information is available17 (e.g. unemployment rate, mark-ups,
interest rates, income and profit tax rates, etc.) or that we want to control (e.g. technological
coefficients, number of agents in each sector, distribution of workers across sectors, loans and capital durations). We thereby obtain initial values for each stock and flow variable of the aggregate
steady state, as well as for some behavioral parameters coherent with the steady/stationary state
(e.g. propensity to consume out of income, target capacity utilization and profit rates, initial capital
and liquidity ratio targets for banks).
4. We distribute each sectors aggregate values uniformly across agents in that sector. In this way
we derive the total value of each type of stock held by agents (e.g. households and firms deposits,
total outstanding loans and real capital for each firm, total loans, reserves and bonds for individual
banks etc.) and agents past values to be used for expectations (e.g. past sales, past wages, past
profits, etc.).
15 For space reasons, we do not present here the exact system of equation representing the steady state. It can be obtained
from the authors, upon request.
16 Notice that the real steady state constraint is due to the fact that the number of households and the technological
coefficients are fixed once and for all.
17 Note that these parameters are set having in mind that every period in the simulation represents a quarter.

16

5. We assume that, in each of the periods before the simulation starts, firms have obtained a loan and
consumption firms have also invested in new capital to maintain their productive capacity constant.
We further assume that the real value (i.e. corrected for inflation) of the new loan or of the new
capital goods was constant in each of these periods. Knowing the constant inflation rate gss and
the amortization schedules for capital goods and loans, we can then derive the outstanding value
for each of these stocks, so that their total value sums up to the amount determined in the previous
step.18
6. In order to set the network structure, we randomly assign a previous period supplier (required for
the matching mechanism) to each demand agent on each market, ensuring that each supplier has
the same number of customers. Similarly, we assign to each single financial stock in households
and firms balance sheets, the asset or liability counterpart by randomly selecting a bank. Again,
we impose that each bank has the same number (and amount) of deposits and loans with the same
number of agents.
The procedure just explained generates an important symmetry condition on agents initial characteristics: that is, we start from a situation of perfect homogeneity between agents in order to limit as
much as possible any possible bias embedded in asymmetric initial conditions, and we let heterogeneity to
emerge from the simulation as consequence of cumulative effects triggered by the stochastic factors embedded in agents adaptive rules. Furthermore, by setting initial values based on SS stock-flow norms we
aim to achieve the threefold objective of limiting our arbitrariness in defining agents initial endowments,
restricting the number of free behavioral parameters in the simulation, and find a consistent criterion to
set the values of several others.
Table 1 in the appendix shows the exact value of the parameters used in the baseline setup, specifying
for each one of them, whether it was exogenously set to determine the steady state (pre-SS), derived
from it (SS-given) or following a logic independent from it (free). Here is a non exhaustive list of the
logic followed to set the values of some important parameters.
The number of workers initially employed by consumption and capital firms, was set to reflect the
relative numerosity of these two classes, so that every firm, regardless the type, starts with the
same number of workers.
The parameters affecting the probability of switching to a new supplier (sec.3.1, were set so that a
15% and 20% difference between the prices of the old and new suppliers gives a probability of 50%
to switch.
Initial mark ups on unit labor costs by consumption and capital firms were set so that the initial
markup on total unit costs is 10 % for consumption and 7.5% for capital firms. The parameters of
the Folded Normal Distribution used in the prices, wages, and interest adaptive strategies were set
so that the expected value of a sample is 0.0075.
The targets used in the growth functions of firms (capacity utilization and return rates) or in the
interest rates setting function of banks (liquidity and capital ratios) were set to their value at the
steady state. Similarly, the macroeconomic threshold used by households to decide whether to
increase or not their asked wage is set equal to the steady state unemployment rate (8%).
Banks risk aversion parameters were computed in order to have a single-period probability of default
1%, given the steady state levels of operating cash flow and debt service costs.19
18 Formally, every firm will have a collection of loans where loan indexed by j = 0, .., 1 was granted j periods before
for an original amount Lj and, taking into account loan amortization, has a current value Lj determined as follows:

Lj = Lj1 (1 + gss )

&

Lj = Lj

&

ot
LT
x0 =

1
X

Lj

i=0

ot
Where LT
x0 is the total debt level of the firm and where is the constant maturity of the loans.
Furthermore, in period 0 each firm will have a collection of batches indexed by i = 0, ..., 1, each one characterized
T ot / and having an accounting value K :
by the same number of machineries k = kx0
j

Kj =

pk0
j
k
,
(1 + gss )j

(5.1)

where pk0 is the price of capital good in the steady state.

19 Note that this value represents the borrowers probability of default estimated by banks for each period from when the
loan is granted to its complete repayment. The estimated overall probability of having a default before the loan is repaid
is higher.

17

 The ratio between banks net worth and banks total assets is set equal to 8%.

Results

After having calibrated the model through the procedure explained in the previous section, we analyzed
the baseline setup by running 100 Monte Carlo simulations for 400 periods. Then, we attempt to validate
the model output by comparing the properties of our artificial time series with their real world counterparts.20 We then describe the dynamics observed in the transition phase of the model and the long term
properties of the economy. We then ran various parameter sweep scenarios, replicating each scenario 5
times. We focus on the parameters affecting (i) consumption firms investment decisions, (ii) banks risk
aversion, and thereby firms access to credit, and (iii) the determinants of firms demand for credit. After
having checked the robustness of our artificial time series fundamental properties across these scenarios,
we finally comment on the impact of these parameters on the model dynamics.

6.1

Validation

In order to validate the model, it is worth comparing, from a qualitative and quantitative point of
view, the properties of artificial and real world times series21 (Assenza et al., 2015). Since our model
is calibrated such that one period corresponds to a quarter, 400 periods correspond to 100 years.Trends
and cycle components have been separated using the Hodrick-Prescott filter.
Figure 2 presents the cyclical component of the logged artificial time series for GDP, unemployment,
investment, and consumption.22 The properties of artificial time series are comparable to observed ones.
The figure shows that investment and unemployment volatility are significantly higher than real GDP
volatility, while consumption is slightly less volatile than output.
In figure 3, we show the artificial and observed auto-correlations of the de-trended series up to the
20th lag. The auto-correlation structure of artificial times series looks remarkably similar to the autocorrelation function of observed real data. All variables have strong positive first order auto-correlations.23
Figure 4 shows the cross-correlations between the cyclical component of real output at time t and of
real output, unemployment, real investment, and real consumption at time t lag. The position of the
peak in each correlation figure indicates whether the variable is lagged, coincident or leading with respect
to output. The shape, dimension and the position of the peak of our simulated time series provide a very
good fit of the properties shown by real time series. Investment and consumption are pro-cyclical and
coincident variables, while unemployment is counter-cyclical and lagged by a quarter.
In appendix B we show how the above highlighted properties of our simulated time series are robust
across the experiments performed in the sensitivity analysis. This suggests the correlation structure
observed in artificial data is an inherent property of the model rather than being dependent on a specific
parametrization. That is, changing the parameters of the model, changes the behavior of agents (thereby
obviously affecting systemic dynamics), but not the underlying correlation structure of macroeconomic
variables. The model thus seems to be able to grasp the structural interdependencies between macro
variables observed in real world.
The model is also able to reproduce several other important stylized facts. Despite agents within each
class being almost perfectly homogeneous at the beginning of our simulations, heterogeneity emerges
during the simulation, first as a consequence of the inherent stochasticity affecting agents interactions
and adaptive behaviors, and then as the result of the path-dependent/cumulative effects arising from
agents competition on real and financial markets.
20 The term validation must be carefully interpreted. Here we generally refer, as usual in the AB macroeconomic literature,
to the model ability to generate micro and macro stylized facts observed in reality. Nonetheless, we are aware of the fact
that this, per se, is not a sufficient condition to accept a theory, or model. The principles of empiricism, dating back to
David Hume, imply the impossibility to validate in any conclusive way any general scientific law, valid in all places and at
all times. This is even more true for Political Economy, given its status of social science. For a detailed discussion on the
epistemological framework of economic theory, see Focardi (2015).
21 Real time series are taken from the Federal Reserve Economic Data (FRED): they are quarterly data ranging from
1955-01-01 to 2013-10-01 for unemployment (not seasonally adjusted, FRED code: LRUN64TTUSQ156N) and ranging
from 1947-01-01 to 2013-10-01 for investments, consumption and GDP (FRED codes: PCECC96, GPDIC96, and GDPC1
respectively).
22 Note the change in scale for the Unemployment graph.
23 A remarkable difference, however, has to be found in the 20th auto-correlation of investment and unemployment which
look significantly higher than real ones. This inconsistency is easily explained in light of our assumption that real capital
has a duration of 20 periods before being scrapped off. This introduces a significant cyclical component in real investment.

18

Unemployment

40

0.4

0.0

0 20

0.4

GDP

100

200

300

400

200

300

400

0.0
0.4

0.4

0.0

0.4

Consumption

0.4

Investment

100

100

200

300

400

100

200

300

400

Figure 2: Cyclical components of logged simulated times series for Output, Unemployment, Investment,
and Consumption

Real Investment

0.5

0.0

0.5 0.0

0.5

0.5

1.0

1.0

Real Output

10

15

20

10

15

Lags

Lags

Unemployment

Real Consumption

20

0.5

0.0

0.5 0.0

0.5

0.5

1.0

1.0

10

15

20

Lags

10

15

20

Lags

Figure 3: Average artificial (continuous) and real (dashed) auto-correlations of the de-trended series up
to the 20th lag. Bars are standard deviations of Monte Carlo average auto-correlations.

In this respect, the selection processes taking place in all markets affects the evolution of their structures. This can be observed in a general tendency towards increasing market concentration. This appears
evident when we look at firms size distribution and banks outstanding credit distribution. Figure 5 high19

Unemployment

0.5

0.0

0.8 0.4

0.5

0.0

1.0

Real Ouptut

10

10

Lags

Lags

Real Investment

Real Consumption

10

0.5

0.4 0.0

0.0

0.4

0.5

0.8

10

10

10

10

Lags

10

Lags

Figure 4: Average artificial (continuous) and real (dashed) cross-correlations of the de-trended series up
to the 10th lag. Bars are standard deviations of Monte Carlo average cross-correlations.

lights that firms and banks are highly heterogeneous with respect to their size distribution. Furthermore,
both the distribution of firms sizes and the distribution of banks credit degree appear to be right
skewed, with upper tails represented by few large firms and banks, having respectively greater dimension
and higher outstanding credit than predicted by a normal distribution.
3P Weibull distribution, AIC = 1696.287

10

20

30
Bin

40

50

0.0000

0.00

0.01

Density
0.02

Density
0.0010 0.0020

0.03

3P Weibull distribution, AIC = 1343.701

60

500

1000

1500
Bin

2000

2500

Figure 5: Panel a: Banks credit degree distribution. Panel b: Firms Dimension

Finally, although our assumptions do not allow for long-lasting real growth, because the number of
workers and the productivity levels are constant and exogenously pre-determined, the model is able to
produce exponential long-term nominal growth and moderate inflation. In the baseline quasi steady-state,
prices of consumption goods and nominal GDP grow at an annual rate of approximately 1.6-1.8% (see
figure 6).

6.2

The transition

The process used for the calibration of our model, as described in the previous section, has to be taken
with a grain of salt. Despite our attempt to constrain our arbitrariness in setting initial conditions, we are
20

aware that the choice of using the logical construct of a specific aggregate steady state with homogeneous
agents and symmetric conditions entails a certain degree of discretionality as well.
Nevertheless, starting from initial conditions derived in such a way does not imply either the model
dynamics stick to the original steady state, nor does it imply the symmetry condition continues to hold
throughout the simulation. Whereas in computing the initial steady state the rate of growth of nominal
variables was fixed exogenously, these constraints are removed as soon as the simulation begins and agents
start to react through their stochastic adaptive rules. As heterogeneity emerges out of the interactions
of the agents composing the model, the inherent dynamics of the model starts appearing.
While in the previous section we focused on the properties of the cycle component of our artificial time
series, this section aims at analyzing the trended dynamics and properties of the model in the baseline
configuration. Therefore, we now use the HP filtered time series focusing on the trend component rather
than on the cyclical one. Results presented in this section are obtained by averaging the trends across
the 100 Monte Carlo simulations ran with the baseline parameter set.
Several forces concur in shaping the dynamics observed in artificial time series. First of all agents
interact in a decentralized way on the different markets. Since the economy lacks of any coordinating
mechanism supervising agents and sectors exchanges, thereby ensuring a balanced growth of inflows and
outflows across sectors, agents autonomous interactions may and generally do lead to significant changes
in the sectoral distribution of real and financial stocks. This in turn, may affect the stability of aggregates
stock-flow norms. When some sectors accumulate increasing liquid resources at the expense of another,
who experiences a drain, we might go into a crisis.
Tracking the evolution of inter-sectoral flows and their impact on sectors balance sheet is thus crucial
to understand the mechanics of the economic system at hand. However, looking at aggregate sectoral
variables is not sufficient to achieve a deep understanding of the model underlying dynamics. A second
type of interaction across agents play a crucial role. This is the interaction between agents within the
same class, which takes the form of the competitive processes undergoing on the different markets.
Therefore, while the analysis of inter-sectoral flows can help identifying aggregate imbalances, the
analysis of the micro evolution of agents balance sheet and market structures may help to identify
imbalances within a sector. For example, it might be the case that while profit margins for consumption
or capital good producers tend to decrease or increase in the aggregate, some firms are recording increasing
profits, while others are experiencing a dramatic deterioration of their financial position, which may result
in a bankruptcy, thereby introducing strong non-linearities in the model.
Interestingly, our artificial time series show the model first experiences a succession of expansionary
and recession trends, then converging, in most cases, to a relatively stable configuration of the economy in
which main real aggregates fluctuate around stable values, and nominal aggregates grow at similar rates,
fluctuating around a steady level. We refer to this situation as a stochastic steady-state, or quasi
steady-state (quasi-SS), while the previous time span constitutes the transition phase of the economy.
Our sensitivity experiments show that the properties of this quasi-SS, such as the level of unemployment, consumption, investment, and inflation, or the presence of cyclicality in trends, depend on the
configuration of agents behavior.
These experiments also show that the convergence to a specific quasi-SS depends on the path experienced by the economy during the transition phase. That is, the long run properties of the economy are
different according to whether agents disperse interactions and adaptive behaviors generate strong intersectoral imbalances over the transition phase or, instead, favor a smooth convergence of stock-flow norms
towards steady, or slightly fluctuating, levels. For this reason we believe it is interesting to understand
the dynamics at stake during the transition phase, before characterizing the quasi-SS of the model.
The transition phase can be divided into three broad phases. Figure 6 presents a dashboard of various
important variables time series.24
Phase 1 - Self-sustained Growth Our simulations starts from a situation of perfect symmetry
between agents within the consumption firms, capital firms, households, and banks sectors. The first
periods of each simulation are thus heavily affected by this particular set-up, showing a rapid increase in
unemployment, followed by an equally rapid recovery.25
24 Growth rates of prices for each simulation have been computed using the average of prices offered by individual firms
in each period, each one weighted for the firms market share.
25 More precisely, at initial conditions, all firms revise the prices of their output in the same direction, leading to a
generalized increase of consumption and capital goods prices. On the contrary, only employed workers revise their asked
wages upwardly, so that average net income do not keep the pace with consumption prices. Households loss of purchasing
power results in a fall of real consumption, and a consequent shrink of employment. However, increased profit margins
stimulate investment. In the meanwhile, as the stochastic variation in firms prices becomes high enough to induce some

21

Growth rates
0.8

Consumption price
Capital price
Net income

0.0

11.0

0.2

11.5

0.4

12.0

0.6

12.5

Nominal GDP (Log)

100

200

300

400

100

200

300

400

33000
32000
30000

2200

Real Investment (left)

Real Consumption (right)

0.05

2000

0.06

31000

0.07

2400

0.08

2600

0.09

0.10

2800

Unemployment rate

100

200

300

400

100

200

300

400

300

400

8000

Constrained Credit

Consumption Firms Funds (Log)

6000

10.0

2000

11.0

4000

12.0

Deposits
Loans

100

200

300

400

100

200

Figure 6: Continuous lines are mean trends over Monte Carlo Simulations. Dashed lines are trends
standard deviations across Monte Carlo runs.

22

The time between periods 10 and 45 displays a self-fuelling and sustained real growth process, mainly
driven by real consumption: higher demand leads to higher employment which, in turn causes further
increases in demand, both directly, and indirectly, through the wages inflationary process undergoing
in a context of low unemployment. Figure 6 shows that during this phase households net-income rises
steadily at a higher (and increasingly) pace than inflation. Demand for consumption goods keeps rising.
On the contrary, real investment in this phase is stable at original values as a consequence of a compensation mechanism between higher capacity utilization rates, positively affecting consumption firms desired
growth, and lower profit rates. Consumption firms profits rates in fact are squeezed by the reduction
of realized markups arising from price and wage dynamics,26 which lowers the numerator, and by the
increased prices of capital goods, augmenting the denominator.
The reduction of profit margins exerts another effect which is crucial in driving the dynamic of the
system, that of reducing the amount of internal funding available to finance production and investment,
relative to the overall financial requirement of firms. The rise of wages and employment increases the
amount of precautionary deposits firms want to hold. This depicts a situation in which, on average,
consumption firms are more reliant on external finance. Credit may help firms to face temporary swings
in their revenues, making up possible short-term shortages of liquid resources. More credit, however, also
means greater amounts of liquid resources being diverted towards debt service. Whether credit acts as
a stabilizing or destabilizing factor primarily depends on which of the two former effects prevails in a
given time span.Figure 6 shows that this tendency is somehow tempered by a tightening of banks credit
rationing behavior, as a consequence of falling operating cash flows.
However, the situation at the micro level is far more variegated. Some firms outperform and experience
increasing profits while some are undergoing a dramatic drain of liquid resources.
Phase 2 - Trend inversion and great recession Wage growth rates tend to stabilize around period
40. Expectations bias on wages seems to fade out, and inflation catches up and surpasses nominal wage
growth, compressing household purchasing power. In the meantime (around period 35) some firms default,
not being able to pay wages and honor their outstanding debt. Firm defaults exerts three effects on the
economic system. First, wages not paid to workers reduce current demand for consumption goods. Second,
fire sales of defaulted firms capital goods reduce household deposits (via the process described in section
4.3). Finally, firms default may result in bad-debt affecting banks balance sheets and capitalization,
with non-trivial effects on firms behavior depending on the dimension of non performing loans, individual
banks exposure, and the number of banks involved.
These processes contribute to an inversion of the expansionary trend and trigger a vicious spiral
augmented by the negative feedbacks between successive slumps in demand and peaks of unemployment.
As a consequence, wages fall and inflation slows down, although the lag characterizing prices dynamics
induces a further loss in household purchasing power. Between period 45 and 130 unemployment rises
fairly steadily, mirroring the progressive fall of real consumption.
Again, the situation at the microeconomic level is more differentiated. Only some of the firms are
able to react to the adverse economic environment and continue to grow, while an increasing number
experiences a deep crisis so that the number of defaults keeps rising. The great recession undergone
by the economy thus affects also the evolution of markets structure, fostering a process of increasing
concentration on real markets.
Phase 3 - Recovery and convergence to the quasi steady-state Around period 130, the dynamics
of the model, which during the previous stages was mainly driven by real consumption, starts to be more
and more dependent on investment. The positive spread between the fall in wages and that in prices, while
reducing households purchasing power, tends to increase consumption firms profit margins. Similarly,
the fall of capital prices reduces firms investment-related outlays and exert a positive impact on the
profit rate (by decreasing the denominator). Finally, defaulted firms are allowed to restart without
the burden of their previous debt, fostering a deleveraging process of the economy. All these aspects
concur to generate a spur of investment, notwithstanding the fall of average capacity utilization rates.
customers to switch towards new suppliers, some firms start reducing their markups in order to recover sales. This dampen
inflation, stimulating a recovery of real wages. The cycle is then reverted.
26 Note that prices and net-income growth rates follow very similar trends, though trends in prices follow the dynamics of
net-income with a lag. This lag is particularly pronounced in the case of consumption prices and can be explained by the
fact that consumption prices are set as an adaptive mark-up over expected unit labor costs. This implies that whenever
wages increase or decrease at increasing pace, simple adaptive expectations systematically underestimate or overestimate
their levels, so that realized mark-ups are systematically lower or higher than desired ones. For expectations to catch up
with actual values, it is required that rates of growth stabilize.

23

Large firms, who are characterized by high rates of capacity utilization despite the general recession,
play a crucial role in fostering the recovery of the economy. While real investment is increasing, real
consumption hits the ground around period 130 and then stabilizes. The system progressively converges
to the quasi steady-state position with unemployment fluctuating just below 8%. After some period, real
investment also stabilizes around the level just sufficient to replace obsolete capital, so that cosumption
firms overall capacity stop growing. It is interesting to notice that the level of real consumption and
real investment in the new steady state are respectively lower and higher than in the intiial situation
while unemloyment is almost the same, implying a redistribution of employees between consumption and
capital firms. In the quasi steady-state neither the aggregate profit rate nor the capacity utilization rate
are at their target levels. Instead, the former is higher while the latter is lower, compensating each other
in the determination of investment.

6.3

Sensitivity analysis

The above discussion highlights the prominence of agents behaviors in shaping the long-term dynamics
of the model. Our previous analysis shows that investment and credit demand behaviors by firms on the
one side, and credit supply behaviors by banks on the other, play a crucial role in steering the economy
across successive expansionary and recessive phases.
Obviously, many other behavioral assumptions may exert an impact on the model dynamics. For
example, different thresholds in workers asked wage function, or different turnover ratios by firms may
turn out to be less favorable to workers, affecting the dynamics of wages and profits. Similarly, tax rates
and other policy parameters characterizing government behavior are likely to exert a huge impact on
the model. Parameters characterizing agents interactions through the common decentralized matching
mechanism may also exert an impact on the process of competition between agents, and consequently on
the evolution of industries and the dynamics of aggregate variables.
Space constraints and explanatory clarity reasons prevent us to perform a sensitivity analysis on all
these parameters. Given their significance in understanding the model dynamics, we chose to focus on
those parameters referring to investment and credit behaviors. We perform 5 sensitivity analysis on the
following parameters: c and k , defining banks risk aversions in assessing consumption and capital firms
reliability, 1 and 2 , defining respectively the weights given to the profit rate and capacity utilization
arguments in consumption firms desired growth function, and expressing firms precautionary deposits
as a share of firms expense for wages. Each sensitivity analysis is carried out by performing a parameter
sweep over the parameter of interest. The range of variation for each parameter is specified in table
1, in the appendix. For each parameter configuration (i.e. for each scenario) we ran 5 Monte Carlo
simulations. Figures 7 and 8 show the averages of the variables trends over the Monte-Carlo simulations
for each sensitivity experiment (each row referring to a specific sensitivity analysis).
6.3.1

Differential risk aversion among banks

In our first sensitivity experiments, we analyze the effects of different risk attitudes by banks in evaluating
borrowers probability of default represented by the parameters k or c in equation (4.15).
The first case, namely the risk aversion of banks when assessing capital firms probability of default,
does not show any significant variation with respect to the baseline scenario. Trends in real consumption,
real investment, unemployment and inflation are overlapping across the different scenarios, suggesting
that capital firms indebtedness plays a minor role in the dynamic of the model.
On the contrary, the first row of figures 7 and 8 show that different degrees of accessibility to credit
by consumption firms exert important and not-trivial effects on the real and nominal dynamics of the
model, both in the medium and long run. High values of banks risk aversion tend to be associated with:
higher real consumption and moderately lower real investment in the long run;
moderately lower unemployment levels in the quasi steady-state;
more volatility in real trends fluctuations in the quasi steady-state;
higher inflation and indebtedness of consumption firms.
The latter result is somewhat surprising, as we would expect more stringent credit constraints by
banks to dampen credit and inflation growth. This apparent paradox can be explained by looking at
the dynamics of the transition phase towards the quasi steady-state: a more prudent attitude by banks
prevents consumption firms to go into too much debt in the first expansionary phase, as well as in the
24

Banks risk aversion towards consumption firms

Real Investment

Unemployment

100

200

300

400

0.09
0.08
0.07
0.06
0.05

2000

30000

2200

31000

2400

32000

2600

33000

0.10

Real Consumption

100

200

300

400

100

200

300

400

300

400

300

400

300

400

Consumption firms growth function: capacity utilization weight

Real Investment

Unemployment

100

200

300

400

0.10
0.05

2000

29000

0.06

2400

0.07

31000

0.08

2800

0.09

33000

3200

Real Consumption

100

200

300

400

100

200

Consumption firms growth function: cash flow weight

Real Investment

100

200

300

400

0.11
0.10
0.05

2000

0.06

2200

0.07

2400

0.08

2600

0.09

2800

33000
32000
31000
30000
29000

Unemployment

3000

Real Consumption

100

200

300

400

100

200

Firms precautionary savings

Real Investment

Unemployment

100

200

300

400

0.09
0.08
0.07
0.06
0.05

2000

30000

2200

31000

2400

32000

2600

33000

2800

0.10

Real Consumption

100

200

300

400

100

200

Figure 7: Consumption, Investment, and Unemployment. Lighter gray lines correspond to higher values
of the parameter.

25

Banks risk aversion towards consumption firms

Consumption Firms Loans (Log)

11.0

11.0

11.5

11.5

12.0

12.0

12.5

12.5

13.0

13.0

GDP (Log)

100

200

300

400

100

200

300

400

Consumption firms growth function: capacity utilization weight

Consumption Firms Loans (Log)

11.0

11.0

11.5

11.5

12.0

12.0

12.5

12.5

13.0

GDP (Log)

100

200

300

400

100

200

300

400

Consumption firms growth function: cash flow weight

Consumption Firms Loans (Log)

11.0

11.0

11.5

11.5

12.0

12.0

12.5

12.5

13.0

13.0

GDP (Log)

100

200

300

400

100

200

300

400

Firms precautionary savings

Consumption Firms Loans (Log)

11.0

11.0

11.5

11.5

12.0

12.0

12.5

12.5

13.0

13.0

GDP (Log)

100

200

300

400

100

200

300

400

Figure 8: Logs of GDP and aggregate outstanding26

debt of consumption firms. Lighter gray lines correspond to higher values of the parameter.

periods of the ensuing recession, lessening in a significant way the amplitude and duration of the recession.
In the two experiments in which the risk aversion parameter is the highest, the economic system is able
to avoid the crisis and shows a smooth transition towards the quasi steady-state. The prevention of the
recession in turn enables the system to attain a balanced growth of prices, wages, profits, investment,
GDP, and credit, avoiding the emergence of excessive imbalances in financial flows between different
sectors. The economy ends up having more credit and more inflation than in the baseline, in spite of the
fact that firms are individually subjected to stricter credit constraints.
6.3.2

Investment behavior

The next parameter sweep consider the case of the weights (1 , 2 ) of the consumption firms desired
growth function (4.11). We start with 2 , the weight given the the rate of capacity utilization. The
second line of figures 7 and 8 highlight that the higher 2 :
the higher real consumption tends to be in the long-run;
the lower investment in the quasi steady-state;
the lower the unemployment rate;
the greater the amplitude of real trends fluctuations in the quasi steady-state.27
The effect on nominal variables, however, are far less trivial. For example, consumption firms indebtedness is significantly higher in the extreme cases (very low or very high value of 2 ). A similar
non-monotonic relationship seem to hold also between 2 and GDP. In fact, nominal output is the highest in the two extreme cases, although the contribution of investment and consumption to GDP differs
between the two cases: nominal investment and nominal consumption are respectively higher and lower,
reflecting the dynamics of their real equivalents discussed above.
In order to understand this non-trivial dynamics we have to look again at the whole path of the economy, from the beginning to the convergence towards the quasi steady-state. In the extreme experiments,
the economy avoids major downturns and this allows a smoother transition phase, which in turn foster
nominal growth, as already explained in the previous section. However, the mechanics by which this
happens is quite different in each of the extreme cases.
In the 0-weight case, investments recover faster during the recession (phase 2 in the transition): real
investment rises faster because of the increasing profit margins arising from capital prices and wage
stagnation (see section 6.2), notwithstanding low demand and capacity utilization rates. The higher
investment in turn compensates for the fall in consumption, lessening the amplitude and duration of the
recession and avoiding major unemployment outbreaks.
On the contrary, in the high-value cases, the investment function dissuade firms to invest too much
when unemployment starts to rise (between phase 1 and phase 2 in transition). This in turn reduces
consumption firms financial requirement, preventing firms from undergoing excess credit. Lower indebtedness implies lower debt service, which increases dividends and reduces the number of defaults relative
to the baseline, fostering again a smoother transition towards the quasi steady-state.
This sensitivity analysis highlights how two opposite mechanisms arising from very different behaviors
by firms may lead to a similar stable configuration of the system.
The second sensitivity experiment considers the case of firms paying more or less attention to their
profitability in deciding investment (1 ). In the zero-weight case, the economy always collapses suggesting
that whenever firms invest based only at their current demand (reflected in the capacity utilization rates),
while disregarding their current profitability, they tend to invest beyond their possibility and default. A
value of 0.005 instead shows a general collapse in 1 out of 5 simulations.
The third row of figures 7 and 8 show the results of our experiments in the non-collapsing cases. We
observe that, as 1 increases:
real consumption and real investment in the quasi steady-state tend to be moderately lower and
higher;
unemployment tends to be slightly lower;
nominal GDP and consumption firms aggregate outstanding debt tend to be lower at the beginning
but higher in the long-run.
27 This can be easily explained in light of the fact that increasing the weight of capacity utilization in the desired growth
function increases the procyclicality of investment.

27

The effects of increasing the weight attributed to the profit rate argument in the desired growth
function of consumption firms, are then somehow similar to the results obtained by increasing the banks
risk aversion and firms capacity weight parameters. Also in this case highest values of 1 prevent a
major crisis, lessening the duration and extent of the recession. However, the mechanics, though similar,
operates with different timing, exerting its effects mainly during the first expansionary phase (phase 1
in the baseline description) rather than in the slowdown and recession phases (phase 2 in the baseline
description): here lower profit margins in the initial phase, arising from the faster increase in wages and
capital prices relatively to consumption prices (see section 6.2), limits firms frenzy in the expansionary
phase preventing them to undergo excessive investment and credit and favoring a smoother transition
towards the quasi steady-state.
6.3.3

Credit Demand

In our final sensitivity experiment, we perform a parameter sweep on the share of expected wage disbursement that firms want to hold as precautionary deposits (). This allows us to investigate the impact
of different credit demand behaviors: the higher , the higher firms credit demand, all other thing being
equal.28 Results of experiments (last row of figures 7 and 8) show that: .
the higher the share of precautionary deposits firms want to hold, the more the system is unstable
and the broader real trends fluctuations tend to be.
Unemployment is significantly lower for the lowest 4 values. Real consumption and real investment
are respectively higher and lower for lower values of ;
long-run nominal GDP and consumption firms indebtedness tends to be higher when firms want
to hold less deposits for precautionary reasons.

Similarly to the banks risk aversion sensitivity analysis, these experiments also give rise to some
counter-intuitive results, as we would expect firms indebtedness and prices inflation to increase the more
firms are prone to borrow. This again can be explained by analyzing the entire path of evolution of
the economy throughout the simulation. Lower values of implies a more conservative behavior of
firms towards credit, lowering debt service and increasing dividends. The consequent smoother transition
towards the quasi steady-state produces similar results to other cases in which major downturns are
avoided. Finally, it is interesting to notice that, while the model is very sensitive to variations of
between 0.5 and 0.9, further progressive increases (0.9 to 1.5) do not change significantly the the long
term properties of the system.

Conclusions

The paper started from a discussion of the potentialities, advantages, and current limits faced by two
innovative modeling approaches to macroeconomics, namely the Agent Based and the Stock Flow Consistent approaches. In that context, we argued in favor of a combination of the two methodologies to be built
on an explicit and realistic modeling of individual agents balance sheets and a consistent representation
of their co-evolution driven by agents disperse exchanges. The present work aims at responding to this
call by providing a relatively simple, general and flexible benchmark macroeconomic AB-SFC model, as
well as a set of coherent rules to calibrate, analyze, and validate it.
Our preliminary results show that the model is able to generate persistent economic fluctuations and
to replicate several important micro and macro stylized facts observed in real economies. The model
provides a good approximation of real world aggregate variables auto and cross correlation structures,
which appears to be robust under different parameterizations, supporting the model candidature to serve
as benchmark.
The analysis of the long term dynamics highlights that, under the majority of cases analyzed, the
economy tends to converge to what we defined as a quasi-SS, thus providing a starting point and a
benchmark to perform policy and scenario analysis in the next works.
28 Before analyzing results it must be stressed that, unlike previous sensitivity analysis in which experiments were conducted on free parameters, parameter is deeply related to the setup of agents initial endowments, as its value was
pre-determined in order to derive the aggregate initial steady state employed to calibrate the model initial conditions.
However, for reasons of comparability between different experiments, we continue to employ the baseline initial setup and
change the values of .

28

In the present work we focused our attention on the analysis of the impact of different investment and
credit behaviors by agents on the transition and the long-term properties of the system, which we carried
out by performing sensitivity analysis on a subset of parameters shaping firms and banks attitudes in
investing and financing.
Results suggest that the interaction of real and financial factors plays a crucial role in driving both
short-term business cycles and the long-term evolution of the economic system. Under certain conditions
firms and banks may be involved in excessive investment and credit, which tends to feed financial and
real instability.
The present work is the starting point of a broader research agenda. In the very next works we would
like to investigate other blocks of the model that, for space and tractability reason, remained on the
background in the present work. The model will then be used for a normative purpose to test different
policy and institutional settings. Among them, some are particularly relevant such as those relating to the
regulation of the labor market for their potential impact on demand and investment patterns. Similarly,
the role of public finance and fiscal policies also deserves more attention. Finally, the adoption of a stockflow consistent framework, providing a coherent and fully integrated picture of the real and financial
sides of the economy, makes our model a suitable tool to assess the impact of monetary, micro-prudential,
macro-prudential policies, as well as the opportunity and efficacy of unconventional policies.
Finally, the framework presented is open to several possible developments and integrations. Some of
them are already ongoing, such as the introduction of R&D investment and innovation dynamics in the
capital good industry, which would open the possibility for long-term, self-sustained real growth in the
model.
The introduction of households indebtedness, real estate markets, and proper portfolio functions by
agents, involving a plurality of financial stocks, such as firms equities, are further examples of already
scheduled integrations to the current framework.

References
Andreasen, M. (2013). Non-linear dsge and the central difference kalman filter. Journal of Applied
Econometrics, 28-6.
Andreasen, M. M., Fernandez-Villaverde, J., and Rubio-Ramrez, J. (2013). The pruned state-space
system for non-linear dsge models: Theory and empirical applications. No. 18983, National Bureau of
Economic Research.
Anufriev, M., Assenza, T., Hommes, C. H., and Massaro, D. (2013). Interest rate rules and macroeconomic
stability under heterogeneous expectations. Macroeconomic Dynamics, 17-8:15741604.
Assenza, T., Delli Gatti, D., and Grazzini, J. (2015). Emergent dynamics of a macroeconomic agent
based model with capital and credit. Journal of Economic Dynamics and Control, 50:528.
Barwell, R. and Burrows, O. (2011). Growing fragilities? Balance Sheets in the Great Moderation.
Financial Stability Paper 10, Bank of England.
Benes, J., Kumhof, M., and Laxton, D. (2014). Financial crises in dsge models: A prototype model. IMF
Working Paper Series, 14/57:59.
Bezemer, D. (2012). The economy as a complex system: the balance sheet dimension. Advances in
Complex Systems, 15:22.
Blanchard, O., Romer, D., Spence, M., and Stiglitz, J., editors (2012). In the Wake of the Crisis: Leading
Economists Reassess Economic Policy. MIT Press , Cambridge MA.
Borogan Aruoba, S., Bocola, L., and Schorfheide, F. (2013). Assessing dsge model nonlinearities. Technical report, NBER working paper series.
Brock, W. and Hommes, C. H. (1997). A rational route to randomness. Econometrica, 65-5:10591096.
Caverzasi, E. and Godin, A. (2015). Post-keynesian stock-flow consistent modeling: A survey. Cambridge
Journal of Economics, 39(1):157187.
Chiarella, C. and He, X. (2001). Asset price and wealth dynamics under heterogeneous expectations.
Quantitative Finance, 1-5:509526.
29

Cincotti, S., Raberto, M., and Teglio, A. (2010). Credit money and macroeconomic instability in the
agent-based model and simulator eurace. Economics - The Open-Access, Open-Assessment E-Journal,
4(26).
Colander, D., Goldberg, M., Haas, A., 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):249267.
Copeland, M. A. (1949). Social Accounting for Moneyflows. The Accounting Review, 24(3):pp. 254264.
Dawid, H., Gemkow, S., Harting, P., Van der Hoog, S., and Neugart, M. (2012). The eurace@ unibi
model: An agent-based macroeconomic model for economic policy analysis.
Dawid, H., Harting, P., and Neugart, M. (2014). Economic convergence: Policy implications from a
heterogeneous agent model. Journal of Economic Dynamics & Control, 44:5480.
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, 204-2:541552.
Delli Gatti, D., Di Guilmi, C., Gaffeo, E., Giulioni, G., Gallegati, M., and Palestrini, A. (2005). A new
approach to business fluctuations: heterogeneous interacting agents, scaling laws and financial fragility.
Journal of Economic Behavior & Organization, 56:489512.
Delli Gatti, D., Gaffeo, E., and Gallegati, M. (2010a). Complex Agent-Based Macroeconomics: a Manifesto for a New Paradigm. Journal of Economic Interaction and Coordination, 5-2:111135.
Delli Gatti, D., Gaffeo, E., Gallegati, M., Giulioni, G., and Palestrini, A. (2008). Emergent Mcroeconomics. An AB Approach to Business Fluctutations. Springer.
Delli Gatti, D., Gallegati, M., Greenwald, B. C., Russo, A., and Stiglitz, J. E. (2010b). The financial
accelerator in an evolving credit network. Journal of Economic Dynamics & Control, 34:16271650.
Dosi, G., Fagiolo, G., Napoletano, M., and Roventini, A. (2013). Income distribution, credit and fiscal
policies in an agent-based keynesian model. Journal of Economic Dynamics and Control, 37-8:1598
1625.
Dosi, G., Fagiolo, G., Napoletano, M., Roventini, A., and Treibich, T. (2015). Fiscal and monetary
policies in complex evolving economies. Journal of Economic Dynamics and Control, 52:166189.
Dosi, G., Fagiolo, G., and Roventini, A. (2010). Schumpeter Meeting Keynes: A Policy-Friendly Model
of Endogenous Growth and Business Cycles. Journal of Economic Dynamics and Control, 34(9):1748
1767.
Esptein, J. (2006). Remarks on the Foundations of Agent-Based Generative Social Science. In Tesfatsion, L. and Judd, K., editors, Handbook of Computational Economics Vol.2, pages 15851602. North
Holland, Amsterdam.
Fagiolo, G. and Roventini, A. (2012). Macroeconomic policy in dsge and agent-based models. Inet
Research Note, -(6):.
Farmer, J. and Foley, D. (2009). The economy needs agent-based modelling. Nature, 460(7256):685686.
Fazzari, S., Hubbard, R., and Petersen, B. (1988). Financing Constraints and Corporate Investment.
Brookings Papers on Economic Activity, 1:141195.
Focardi, S. M. (2015). Is economics an empirical science? if not, can it become one. Name: Frontiers in
Applied Mathematics and Statistics, 1(5).
Godley, W. (1997). Macroeconomics without Equilibrium or Disequilibrium. Working Paper Series 205,
The Levy Economic Institute of Bard College.
Godley, W. and Cripps, F. (1983). Macroeconomics. Oxford University Press.
Godley, W. and Lavoie, M. (2007). Monetary Economics An Integrated Approach to Credit, Money,
Income, Production and Wealth. Palgrave MacMillan, New York.

30

Godley, W. and Wray, L. R. (1999). Can goldilocks survive?

Institute of Bard College.

Technical report, The Levy Economic

Godley, W. and Zezza, G. (2006). Debt and lending: A cri de coeur. Technical report, Levy Institute at
Bard College.
Greenwald, B. C. and Stiglitz, J. E. (1993). Financial market imperfections and business cycles. The
Quarterly Journal of Economics, 108(1):77114.
Kinsella, S. (2011). Words to the wise: Stock flow consistent modeling of financial instability. Available
at SSRN 1955613.
Kinsella, S., Greiff, M., and Nell, E. J. (2011). Income Distribution in a Stock-Flow Consistent Model
with Education and Technological Change. Eastern Economic Journal, 37:134149.
Lavoie, M. (1992). Foundations of Post-Keynesian Economic Analysis. Edward Elgar, Aldershot.
McLeay, M., Radia, A., and Thomas, R. (2014). Money creation in the modern economy. Bank of England
Quarterly Bulletin, Q1:14.
Meyers, S. (1984). Capital Structure Puzzle. Journal of Finance, 39-3:575592.
Raberto, M., Teglio, A., and Cincotti, S. (2012). Debt, deleveraging and business cycles: An agent-based
perspective. Economics - The Open-Access, Open- Assessment E-Journal, 27:50.
Riccetti, L., Russo, A., and Gallegati, M. (2014). An agent-based decentralized matching macroeconomic
model. Journal of Economic Interaction and Coordination, 3:.
Seppecher, P. (2012). Monnaie Endog`ene et Agents Heterog`enes dans un Mod`ele Stock-Flux Coherent.
In Political economy and the Outlook for Capitalism, Paris (France).
Simon, H. (1955). A Behavioral Model of Rational Choice. The Quarterly Journal of Economics, 691:99118.
Simon, H. (1976). From Substantive to Procedural Rationality. In Latsis, S., editor, Method and Appraisal
in Economics, pages 129148. Cambridge University Press, Cambridge UK.
Steindl, J. (1952). Maturity and Stagnation in American Capitalism. Blackwell.
van der Hoog, S. and Dawid, H. (2015). Bubbles, crashes and the financial cycle: Insights from a stockflow consistent agent-based macroeconomic model. Working Papers in Economics and ManagementUniversity of Bielefeld, 1-2015:58.

31

Parameters values and Initial Setup

Table 1: Parameters
Symbol
gSS : pre-SS
sizeH : pre-SS
sizeC : pre-SS
sizeK : pre-SS
sizeB : pre-SS
Ngt : SS-given
Nc0 : pre-SS
Nk0 : pre-SS
u0 : pre-SS
N : pre-SS
{k , lk }: pre-SS
c = k : free
d = l : free
n : free
d = l : free
c = k : free
: pre-SS
: free
: free
c0 : pre-SS
k0 : pre-SS
(F N , F2 N ): free
: pre-SS
= i : pre-SS
: pre-SS
: pre-SS
r: SS-given
u: SS-given
1 : free
2 : free
: pre-SS
c = k : pre-SS
b : pre-SS:
ilb0 : pre-SS
idb0 : pre-SS
CR0T : SS-given
LR0T : SS-given
c : free
k : free
a
icb : pre-SS

Description
Nominal rate of growth in the SS
Number of households
Number of consumption firms
Number of capital firms
Number of banks
Number of public servants (constant)
Consumption firms initial workers
Capital firms initial workers
Initial unemployment
Productivity of labor in K sector
Productivity and capital/labor ratios of
K
Number of potential partners on C and
K goods mkts
Number of potential partners on
deposit-credit mkts
Number of potential partners on labor
mkt (for each vacant job)
Intensity of choice in deposit-credit
mkts
Intensity of choice in C and K goods
mkts
Firms inventories target share
Adaptive expectations parameter
Labor turnover ratio
Initial mark-up on ULC for C firms
Initial mark-up on ULC for K firms
Folded Normal Distribution parameters
Profit tax rate
Profit and Income tax rates
Loans duration
Capital goods duration
Target profit rate (Investment function)
Target capacity utilization (Investment
function)
Profit rate weight (Investment function)
Capacity utilization rate weight (Investment function)
Firms precautionary deposits as share
of WB
Firms profits share distributed as dividends
Banks profit share distributed as dividends
Initial interest rate on loans
Initial interest rate on deposits
Initial banks target capital ratio
Initial banks target liquidity ratio
Banks risk aversion towards C firms
Banks risk aversion towards K firms
CB interest rates on advances

32

Baseline
0.0075
8000
100
20
10
1360
4000
1000
0.08
2
{1, 6.4}

Sensitivity
same
same
same
same
same
same
same
same
same
same
same

same

same

10

same

2.00687

same

1.50515

same

0.1
0.25
0.05
0.318857
0.075
(1,0.0094)
0.18
0.18
20
20
0.04345
0.8

same
same
same
same
same
same
same
same
same
same
same
same

0.01

0.000 : 0.005 : 0.035

0.02

0.000 : 0.005 : 0.040

0.5 : 0.1 : 1.5

0.9

same

0.6

same

0.0075
0.0025
0.17996
0.258026342
3.92245
21.51335
0.005

same
same
same
same
1.0 : 1.0 : 8.0
5.0 : 5.0 : 40.0
same

Table 1: Parameters
Symbol
: free

Description
Haircut on defaulted firms capital
value
Initial wages
Dole (share of average wages)
Unemployment threshold in wage revision function
Propensity to consume out of income
Propensity to consume out of wealth
Bonds interest rate
Bonds price

wn0 : pre-SS
: pre-SS
: free
1 : SS-given
2 : pre-SS
b
i : pre-SS
pb : pre-SS

Households

Cons. Firms
CA

KA

32971.4

+32971.4

+36800

25000

+1280

CG on inventories

Investments
Capital Amortization

Consumption

Sensitivity
same

5
0.4
0.08

same
same
same

0.38581
0.25
0.0025
1

same
same
same
same

Banks

Govt.

CA

KA

CA

KA

5000

+22.3

22.3

+3.7

3.7

5375

+5375

4974

+4974

7084.7

484.8

68.7

Wages
Dole

Taxes

Cap. Firms

Baseline
0.5

Dep. Interest

Central Bank

CA

KA

6800

-1280

39.3

+7677.4

+200.3

+62

+12.4

274.7

Bonds Interest

+95

165.9

+70.9

Loans Interest

388.5

9.7

+398.2

Advances Interest

+2367.6

2208.4

+220.8

312.8

+31.3

179.1

+71.7

CB profits

+70.9

70.9

Cash

Deposits

600.8

186.1

37.2

+824.1

Advances

Reserves

212.6

+212.6

Gov. Bonds

284.9

+497.6

212.6

Loans

+388.5

+9.7

398.2

Total

Profits

Table 2: Aggregate Transaction Flow Matrix (Initial Situation)

Households

Cons. Firms

Cap. Firms

Banks

Govt.

Central Bank

Total

+80704.1

+25000

+5000

110704

Loans

52194.4

1298

+53492.5

Cons. Goods

+2997.4

+2997.4

Cap. Goods

+53863.6

+500

+54363.6

Bonds

+38273.5

66838.1

+28564.6

Reserves

+28564.6

28564.6

Advances

+80704.1

+29666.6

+4202

+9626.4

66838.1

+57361

Cash
Deposits

Net Worth

Table 3: Aggregate Balance Sheet (Initial Situation)

Auto and cross correlations in our experiments

33

Artificial Autocorrelations

Artificial Crosscorrelations
Real Ouptut

20

10

15

20

10

15

20

0.4
0

10

10

15

20

10

1.0
0.5

0.2
0.2
0.6
0

10

Real Consumption

0.6

1.0
0.0
0.5
5

Real Investment

0.5

0.5
0.0
0.5
0

0.4
10

Real Consumption

1.0

Real Investment

0.0

15

0.5

10

0.8

0.0
0.5

0.0
0.5
5

0.0

0.5

1.0
0.5

0.5
0.0
0.5
0

Unemployment

1.0

Unemployment

1.0

Real Output

10

10

10

10

Figure 9: Auto (left) and Cross (right) Correlation in the Banks risk aversion scenario. Data are average
correlations and their average standard deviations across Monte Carlo runs for each scenario.

Artificial Autocorrelations

Artificial Crosscorrelations
Unemployment

Real Ouptut

Unemployment

20

0.0
0

10

15

20

Real Consumption

10

15

20

10

0.6
0.2
0.6
0

10

15

10

20

10

Real Consumption

0.2

0.0
0.5
5

Real Investment

0.5

1.0
0.5
0.0
0.5
0

0.8
10

1.0

Real Investment

0.8

15

0.4

10

0.0

0.4

0.4

0.4
0.0
0.4

0.5

0.5

0.0

0.0

0.5

0.5

0.8

1.0

1.0

Real Output

10

10

10

10

Figure 10: Auto (left) and Cross (right) Correlation in the Firms capacity utilization weight scenario.
Data are average correlations and their average standard deviations across Monte Carlo runs for each
scenario.

34

Artificial Autocorrelations

Artificial Crosscorrelations
Real Ouptut

Unemployment

20

10

15

20

1.0
20

10

0.5
0.5
0

10

15

10

20

10

Real Consumption

0.0

0.0
0.5
15

Real Investment

0.5

1.0
0.5
0.0

10

0.6

Real Consumption

0.5

1.0
10

1.0

Real Investment

0.2

0.5
0.0
0

1.0

15

0.5

10

0.0

0.5

0.5

0.5

0.5

0.0

0.0

0.5

0.5

0.2

1.0

1.0

Unemployment

1.0

Real Output

10

10

10

10

Figure 11: Auto (left) and Cross (right) Correlation in the Profit rate weight scenario. Data are average
correlations and their average standard deviations across Monte Carlo runs for each scenario.

Artificial Autocorrelations

Artificial Crosscorrelations

20

10

15

20

20

10

0.0
0.5
0

10

15

10

20

10

Real Consumption

0.5

1.0
0.0
0.5
15

Real Investment

0.5

1.0
0.5
0.0

10

0.5

Real Consumption

0.5

1.0
10

1.0

Real Investment

0.0

1.0
0.5
0.0
0

1.0

15

0.5

10

0.0

0.5

Unemployment

0.5

0.5

0.5

0.0

0.0

0.5

0.5

1.0

Real Ouptut
0.5

Unemployment

1.0

Real Output

10

10

10

10

Figure 12: Auto (left) and Cross (right) Correlation in the Precautionary deposits scenario. Data are
average correlations and their average standard deviations across Monte Carlo runs for each scenario.

35