Technological Forecasting & Social Change · 2017. 10. 16. · Technological Forecasting & Social...

Technological Forecasting & Social Change xxx (2016) xxx–xxx

TFS-18613; No of Pages 19

Contents lists available at ScienceDirect

Technological Forecasting & Social Change

Integrated crisis-energy policy: Macro-evolutionary modelling of technology, financeand energy interactions

Karolina Safarzyńska a,⁎, Jeroen C.J.M. van den Bergh b,c,1

a Faculty of Economic Sciences, Warsaw University, Dluga 44/50, 00-241 Warsaw, Polandb ICREA, Barcelona, Spainc Institute for Environmental Science and Technology, Universitat Autònoma de Barcelona, Edifici Z — Campus UAB, Bellaterra, 08193, Spain

⁎ Corresponding author.E-mail addresses: [email protected] (K. Saf

(J.C.J.M. van den Bergh).1 Also affiliated with Faculty of Economics and Busines

for Environmental Studies, VU University AmsterdamTinbergen Institute.

http://dx.doi.org/10.1016/j.techfore.2016.07.0330040-1625/© 2016 Elsevier Inc. All rights reserved.

Please cite this article as: Safarzyńska, K., vafinance and energy interactions, Technol. Fo

a b s t r a c t
a r t i c l e i n f o
Article history:Received 1 December 2015Received in revised form 6 July 2016Accepted 22 July 2016Available online xxxx

Addressing four persistent problems, namely human-induced environmental change, financial instability, in-equality and unemployment has now become an urgent necessity. To better grasp complex interactions betweentechnological, financial and energy systems, we propose a formal behavioral-evolutionary macroeconomicmodel. It describes the coevolution of four populations, namely of heterogeneous consumers, producers,power plants and banks, interacting through interconnected networks. We examine how decisions by all theseeconomic agents affect financial stability, the direction of technological change and energy use. The approachgenerates non-trivial, even surprising insights, such as that brand loyalty, captured by a network externality onthe demand side, can increase the likelihood of bankruptcies of banks. Cascades of such bankruptcies are foundto bemore likely under greater income inequalities and higher electricity prices. We employ themodel to assessmacroeconomic impacts of sustainability policies along three dimensions: environmental effectiveness, financialstability and socio-economic consequences.

© 2016 Elsevier Inc. All rights reserved.

Keywords:Financial-economic crisisEnergyInequalityMacroevolutionary modelling

1. Introduction

Theworld's inability to solve current persistent problems of an envi-ronmental, financial and socio-economic nature has now become fla-grant. Much has been written about potential solutions to stabilize thefinancial system, to solve environmental problems, or to address persis-tent unemployment and inequality. However, most existing studieslimit themselves to only one of these problems, without paying atten-tion to how different subsystems interact or how solving one problemwould affect the other challenges. Nevertheless, there are many indica-tions now that these problems are not independent but intricately con-nected. While their isolated study greatly simplifies formal analysis, itignores relevant feedback mechanisms underlying the ultimate eco-nomic dynamics. Moreover, it maymean that important causes of prob-lems or crises may be downplayed or even overlooked, such as the roleof energy scarcity and prices in the outbreak of the last financial crisis,and hence certain potential solutions remain unnoticed. In addition,the role of social interactions, such as imitation or diffusion of

arzyńska), [email protected]

s Administration, and Institute, The Netherlands. Fellow of

n den Bergh, J.C.J.M., Integratrecast. Soc. Change (2016), ht

knowledge, in causing instability in financial systems has been fully ac-cepted in behavioral finance but is still not well integrated in traditionalmacroeconomic approaches. Not surprisingly, from many corners it isnow argued that a new approach tomacroeconomics is needed for a ro-bust analysis of multiple relevant policies (Farmer and Foley, 2009;Schweitzer et al., 2009; Stiglitz and Gallegati, 2011).

To this end, we propose a general model for studying feedbackmechanisms between finance, technology and energy systems, to sub-sequently derive lessons for sustainability policies. The proposed frame-work consists of populations of heterogeneous consumers, producers,power plants and banks interacting through interconnected networks.The starting point of our analysis is thatwithout a proper understandingof the fundamental relationships between sub-systems in the economy,our ability to guide their interactions towardsmore economically stableand environmentally sustainable trajectories is unnecessarily limited.Worse, policies suggested for one subsystem may rebound throughtheir indirect impacts on the other subsystems, resulting in a failure ofmeeting all desired goals associated with each subsystem. Currentmodels dealing with each issue separately are, however, incapable ofidentifying – let alone quantifying – such indirect effects. As a result,they may overestimate the effectiveness of various policies, and possi-bly even contribute to promotion of alleged policy solutions that canturn out to be erroneous. The proposed new approach allows examina-tion of interactions between policies in different domains, which aids indesigning effective policy packages.

ed crisis-energy policy: Macro-evolutionary modelling of technology,tp://dx.doi.org/10.1016/j.techfore.2016.07.033

http://dx.doi.org/10.1016/j.techfore.2016.07.033

mailto:[email protected]

Journal logo


http://www.sciencedirect.com/science/journal/00401625


2 K. Safarzyńska, J.C.J.M. van den Bergh / Technological Forecasting & Social Change xxx (2016) xxx–xxx

Attention for social networks inmacroeconomic analysis is importantas well, as crises often take the form of cascades of failures spreadingthrough financial networks (Battiston et al., 2012; Acemoglu et al.,2013; Tedeschi et al., 2012; Thurner and Polenda, 2013), or through net-works of firms producing final and intermediary products (Henriet et al.,2012). Typically, macroeconomic agent-based models, which explainbusiness cycles as emerging from firms' innovative activities, eitheromit the banking sector entirely (as discussed in Section 2) or the bank-ing sector is reduced to one bank (e.g. Battiston et al., 2007; Assenzaet al., 2015). On the other hand,models of contagion in financialmarketsrely on a simplified description of firms, focusing on firms' investmentsin risky assets while ignoring their core production activities (Tedeschiet al., 2012; Thurner and Polenda, 2013; Gabbi et al., 2015). These ap-proaches do not allow studying how production factors and associateproblems of energy use (scare fossil fuels or carbon pricing) and labor(employment) relate to financial processes and public regulation. Onlyfew agent-based models make a connection between networks of het-erogeneous firms and banks (Cincotti et al., 2010; Gaffeo et al., 2008;Delli Gatti et al., 2009; Neveu, 2013). Our approach fits in this line of re-search. The novelty of ourmodel is that it combines financial and energymarkets, and bridges macro-evolutionary (e.g., Delli Gatti et al., 2011;Fagiolo and Roventini, 2012, 2016) and environment-evolutionarymodeling (Safarzynska et al., 2012). This results in a new approach thatallows for studying integrated crisis and environmental policies in rigor-ous and systemic manner.

The approach also ideally allows to incorporate insights frombehav-ioral economics. Akerlof (2002) strongly advocated so-called “behavior-al-macroeconomics”. Models of habits formation and social imitationhave been shown to have important macroeconomic implications. Forinstance, habit formation model explains the excess smoothness of ag-gregate consumption (Campbell and Cochrane, 1995; Lettau andUhlig, 2000; Dynan, 2000). A study by Ljungqvist and Uhlig (2000)showed that if keeping up with Joneses, agents tend to overconsume.This can be corrected with tax policy, namely through countercyclicaleconomic effects provoked byprocyclical taxes: “cooling down” theeconomy with higher taxes when it is “overheating” in booms and“stimulating” the economy with lower taxes during recessions to keepconsumption up. Behavioral utility functions, notably interdepen-dencies between the utility of multiple individuals, often imply thatthe consumption or social problem cannot be treated as a straightfor-ward dynamic optimization problem (Laibson, 1998). Other ap-proaches, such as agent-based models, offer a powerful tool then tostudy associated macroeconomic phenomena.

The model proposed here builds upon our precedingwork, in whichwe modeled technological change as resulting from interactions be-tween heterogeneous consumers interested in new products andfirms undertaking product and process innovations (Safarzynska andvan den Bergh, 2010a,b). This model was extended by integrating elec-tricity produced from diverse energy sources as an important factor ofproduction in Safarzynska and van den Bergh (2011) and Safarzynska(2012). Here, we further extend thesemodels by adding a financial sec-tor as well as labor markets. In the market for consumer goods, atechnological trajectory arises from the interplay of incremental innova-tion and the search for new product designs by individual firms, follow-ing the seminal work by Nelson and Winter (1982). New firms asks abank for a startup loan. This way firm-bank inter-credit linkages evolveas a result of activities in the real economy. If banks have no sufficientliquidity, they ask other banks for loans in the interbank lending net-work. In modeling the interbank lending market, we follow Thurnerand Polenda (2013).

In our model, electricity is assumed to be an important input in theproduction of consumer goods, alongwith capital and labor. This is illus-trated by electricity being essential to manufacturing: it can reach up to95% of total energy use for production (Steinbucks, 2010). The electric-ity market is modelled as composed of heterogeneous plants producingelectricity from diverse energy sources. Over time, new power stations

Please cite this article as: Safarzyńska, K., van den Bergh, J.C.J.M., Integratfinance and energy interactions, Technol. Forecast. Soc. Change (2016), ht

enter the market. The discounted value of investments in each energytechnology determines the type and size of power plants to be installed.

On the demand side, consumers imitate choices of others withintheir social networks. We distinguish three consumer classes based onthe source of their income, namely owners of the factors capital and en-ergy, and workers. Energy owners can be thought of as shareholders ofenergy utilities or companies working in oil and gas, while capitalownersmay be regarded as small producers who own somemachinery.In our model, consumers evaluate the attractiveness of different prod-ucts based onwhether others in their socio-economic class have alreadyadopted them. The stronger the brand loyalty is, the more likely theclustering of consumer choices occur around similar products. Onemay think of products in our model as electronics, as here brand loyaltyand social comparisons play an important role. For instance, the pur-chase of smartphones is strongly influenced by brand loyalty. In fact,many consumers frequently upgrade their phone merely to keep upwith the Joneses. This partly explains why the smartphone market isso innovative, showing new products to appear every year and stronglycompeting for adoption. Distinguishing three heterogeneous consumerclasses allows us to study the impact of distributive policies on financialand economic stability beyond other agent-based models. Consumerdifferentiation is behind arising income inequalities, while the degreeto which consumers imitate others in their social networks determinesthe degree of market competition.

Our model for the first time conceptualizes connections between fi-nance, energy and labor. It allows us to examine the potential trade-offsbetween financial and economic stability. We show that if the networkeffect on the consumer side is weak, the more likely bankruptcies ofbanks are to occur. This is because under a weak network effect, themarket resembles the fashion market with many firms competing foradoption. As a result, more firms ask banks for loans, which translatesinto a higher connectivity in the interbank lending market. On theother hand, conventional indicators, used by banks to evaluate riskinessof firms' projects, are less informative here compared to markets char-acterized by a strong network effect. In addition, the cascades ofbanks' failures are more likely the greater are income inequalities andthe higher is the price of electricity. In particular, high income inequal-ities lead to wealth accumulation by some socio-economic groups,undermining demand. In addition, we find that high prices of electricitydrive up the total debt in the economy, causing inflation. Finally, we em-ploy the model to assess macroeconomic impacts of sustainability poli-cies along three dimensions: environmental effectiveness, financialstability and socio-economic consequences. Specific instruments con-sidered include: distributive policy; renewable energy subsidies, andregulations of bank lending to firms.

The reminder of this paper is organized as follows. Section 2 explainsthe general model approach and its connection with the existing litera-ture. Section 3 discusses the detailed model structure. Section 4 exam-ines the potential trade-offs between financial and economic stability.Section 5 discusses policy lessons derived fromourmodel for suitability.Section 6 concludes.

2. General approach

Our study connects with the theoretical literature on agent-basedmodels that combine networks of heterogeneous firms and banks(Cincotti et al., 2010; Gaffeo et al., 2008; Delli Gatti et al., 2009; Neveu,2013) and coevolutionary models of supply and demand dynamics(Janssen and Jager, 2002; Windrum and Birchenhall, 1998, 2005; Oltraand Saint-Jean, 2005; Saint-Jean, 2006; Windrum et al., 2009a,b;Malerba et al., 1999, 2001, 2009; Safarzynska and van den Bergh,2010a,b). Agent-based models combing financial markets with thereal economy have shown that the cascades of bankruptcies can propa-gate between networks of heterogeneous firms and banks. In suchmodels, a bankruptcy of one firm can trigger bankruptcies in intercon-nected firms or banks. On the other hand, coevolutionary models of



3K. Safarzyńska, J.C.J.M. van den Bergh / Technological Forecasting & Social Change xxx (2016) xxx–xxx

supply and demand have generated novel insights about how to unlocka market dominated by an unsustainable technology and how to effec-tively stimulate diffusion of “green” innovations. Here, technologicalchange results from interactions within and between consumer andproducer populations. On the supply side, firms engage in a search pro-cess for better technological solutions. In particular, firms can improvetheir total factor productivity by undertaking R&D activities or by copy-ing the industry's best practices. This approach follows the seminalwork by Nelson and Winter (1982). Consumers are boundedly rationaland imitate otherswithin their networks. Their preferences change overtime as a result of social interactions, which then co-determines the di-rection of technological change. Typically, distinct consumer classes aremodeled, which explains the emergence of distinct market niches withfirms specialized in satisfying needs of particular consumer classes (e.g.Windrum and Birchenhall, 2005; Safarzynska and van den Bergh,2010a,b). This approach is now quite common in agent-based modelswith heterogeneous consumers and producers.

Our approach to modeling the interbank market builds on recentnetwork models of financial markets (Farmer and Lo, 1999; Gai, 2013;Thurner and Polenda, 2013). Typically, banks are modeled here asnodes in a financial network. Such a representation has proved usefulfor studying the impact of network topology on systemic risk. For in-stance, Thurner and Polenda (2013) assess risk associated with lendingto agents in various nodes of the financial network by using a metric oftheir interbank liability network. They show that systemic risk in finan-cial networks can be drastically reduced by increasing transparency, i.e.making the networkmetrics of individual banks visible to others, and byimposing a rule that prevents interbank borrowing from banks occupy-ing nodes associated with a high degree of systemic risk. Similarly, in amodel by Tedeschi et al. (2012), liquidity problems of banks can spreadthrough the network of banks. In this framework, companies make in-vestments using the credit market. The authors explore the relationshipbetween cascades of bankruptcies and business cycles. They find thatincreasing interbank connectivity increases the risk of cascades ofbank failures. In the presence of high reserves requirements, an in-creased interbank connectivity obstructs economic growth because lit-tle credit is available to companies.

Fig. 1 presents a schematic structure ofmodules and interactions be-tween energy, labor and financial sectors. In our model, we consider aproduct market with many firms producing highly differentiatedgoods. Contrary to preceding studies on demand-supply coevolution,

Fig. 1. Schematic struc


here we distinguish three consumer classes based on the source oftheir income, namely owners of the factors capital and energy, andworkers. Workers sell their labor to firms producing final goods andsubsequently spend their income on purchasing products. The rent re-ceived by capital and energy owners depends on total expenses byfirms on capital expansion and energy. In particular, total amountsspent on energy and capital are distributed equally among energy andcapital owners, respectively. Our model explicitly focuses on the inter-dependence of preferences of consumers through their social networks.In particular, the utility derived by a consumer from purchasing a gooddepends on interpersonal comparisons and relative positions. The rec-ognition of the interdependence of utility functions goes back toDuesenberry (1949) and Veblen (1922), who argued that social com-parison and status-seeking are important determinants of individualbehavior. In line with this, we assume capital and energy owners wantto avoid purchasing products that are commonly bought by workers.This reflects a desire to distinguish oneself from the majority throughthe purchase of status commodities, referred to in the literature as thesnob effect (Leibenstein, 1950). We chose capital and energy ownersas a group which tries to distinguish themselves from workers, as theyenjoy higher incomes than workers, and thus they can afford buyingmore expensive products.

On the supply side,firmsdecide about a desirable level of productionand the associated use of inputs (labor, capital and electricity). Productquality changes over time as a result of learning-by-doing (experience).The effect of incremental improvements in product design on sales isuncertain. As a result of a change in product quality a firm can attractnew consumers, while it may lose others. In addition, every periodone new firm offering a new product design tries to enter the market.It asks a randomly chosen bank for a start-up loan. Only if a bank hassufficient liquidity, or is able to raise capital in the interbank (IB) lendingmarket, the firm enters the market. Similarly, incumbent firms can asktheir banks for loans, for instance, to invest in capital expansion. Theloans are granted on two conditions: (1) a bank has a sufficient liquidityor is capable of raising finance in the interbank lending market and(2) the debt-to-equity ratio of a firm asking for a loan does not exceedsome critical value.

Firms employ electricity as an input in production, which is pro-duced by energy companies. Initially, three plants produce electricity.Each plant employs a different fuel in production. We distinguishthree stylized fuels, characterized by constant fuel costs; costs which

ture of the model.




increase over time; and costswhichdecrease over time. A plant is closedafter reaching itsmaximum lifespan, and a newplant enters themarket.The size of a new power plant and the type of energy technology, whichit employs, are chosen based on the discounted value of future invest-ments. The new plant receives a loan from a bank to cover initial invest-ments in the newly installed capacity, which it has to pay back afterreaching its maximum lifespan.

The bankruptcy of a bank can arise due to bankruptcy of firms orother banks, and consequent failures of loan repayments (Thurnerand Polenda, 2013). If a bank goes bankrupt, its loans at the inter-bank market are written off, while consumers' and firms' savingsare reallocated to a randomly chosen other bank. In case reservesof banks are insufficient to pay back all deposits, savings of banks' cli-ents may be reduced if necessary so that all consumers lose the sameshare of their deposits. On the other hand, loans of firms (of powerplants and producers of final goods) are reallocated to other banks,to which those firms are randomly assigned. We do not consider en-tries of new banks. This is a simplifying assumption. In reality, bankstake over one another when one fails and consolidate. Alternatively,central banks act as a lender of last resort so as to rescue banks at theedge of bankruptcy. Yet, as our aim is to understand the conditionsunder which banks fail, we do not model these processes explicitly.We ensure that the monetary base in the economy remain constantthroughout model simulations.2

In themodel, networks of consumers, firms, power plants and banksare linked via numerous feedback mechanisms, which can cause thespread of systemic risk between them. For instance, the bankruptcy ofa firm can render its bank insolvent. In turn, the lattermay trigger an av-alanche of bankruptcies in the interbank lending network. In addition,the collapse of any firm causes an immediate fall in employment, in-creasing the probability of bankruptcies of other firms via diminishedpurchasing power of laid-off workers.

Savings in the economy affect the probability of crises. In particular,they determine investment opportunities by banks, while the lack ofsavingsmay curb economic growth by undermining consumer purchas-ing power. Savings due to profit accumulation by successful firms canhave two opposing effects on the economy. On the one hand, firms' sav-ings (i.e. deposits at the bank) allow banks to financemore innovations,which typically render new job opportunities. On the other hand, theycan harm economic growth and employment: undermonopolistic com-petition, firms are able to impose high markups on their products, andaccumulate profits over time, of which only a small percentage comesback to workers as wages. This may diminish purchasing power of con-sumers, and as a result chances of new firms being successful. Ultimate-ly, if unsuccessful firms are unable to pay back their loans this may leadto a cascade of bankruptcies of banks.

3. Technical specification of the model

We consider a market economy populated by a large number offirms {1,…, nFt} banks {1,…, nb} power plants {1,…, np}, and consumers{1,…, nc}, who undertake decisions at discrete time, denoted by t=0, 1,2,…, T. In each time period, the following sequence of steps is repeated:

1) Each consumer attempts to purchase a product that provides thehighest utility: she (implicitly) ranks all offers and attempts to buythe most attractive product given his budget constraints (savingsdeposited at the bank). If the product is out of stock, she attemptsto buy the second best offer.

2 In rare cases, total reserves increase in some model simulations by a small amount.This is due to unforeseen bank losses, such as simultaneous bankruptcies of banks andfirms. This effect has, however, been negligible in the simulations and did not affect theresults.


2) Firms collect profits and set the desired production level for the nextperiod as aweighted average between past sales and the desired de-mand. They employ labor and electricity in accordance with this.They also undertake R&D activities. Formally, firms sample produc-tivity from a distribution of technological opportunities.

3) If the desired production level exceeds the firm's production capac-ity, a firm invests in capital expansion. A firm asks its bank for a loanif its profits are insufficient to cover the necessary expenses. A banklendsmoney to the firm if it has a sufficient capital, otherwise it bor-rows money in the interbank lending market, or no loan is paid out.

4) The power plants decide howmuch electricity to produce given theexpected demand for energy from firms. If an incumbent plantreaches it maximum lifespan, a new plant enters the market. Itssize and the type of energy technology which it embodies dependon the discounted value of investments. A new plant receives aloan from the bank to cover expenses in its installed capacity.

5) The number of firms on the market changes over time as a result ofentry and exit of firms. Each time step, a new firm enters themarketwith some probability. It asks a randomly chosen bank for the start-up funding, and enters themarket conditional on receiving the loan.The newcomer offers a product whose quality may outperform thatof incumbent firms.

6) A firm leaves themarket if it goes bankrupt and it is not able to repayits debts.

7) Banks repay loans in the interbankmarket. If a bank's reserves or eq-uity become negative, the bank goes bankrupt. Negative reservesoccur if banks are unable to pay their outstanding debts. Equity iscalculated by subtracting liabilities from bank's assets. It can becomenegative as a result of bankruptcies of firms.

In Section 3.1, we discuss specific assumptions made about firms,in Section 3.2 about consumer behavior, in Section 3.3 about banksin the interbank lending market, and in Section 3.4 about powerplants.

3.1. Firms in manufacturing industry

The model describes nf firms producing a homogenous, but highlydifferentiated, product. Each firm j offers a single good, with a design,denoted by xjt, that is randomly sampled from the range (0,~xt) at the be-ginning of each simulation. Here,~xt ¼ ς~xmaxt is themaximum attainablequality at time t, with ς being a positive fraction of themaximumquality~xmaxt that firms can achieve in time t = 0.

A firm j sets a target level of production for the next periodas a weighted average of its current sales sjt and actual demand djt(cf. Windrum and Birchenhall, 1998, 2005):

~yjtþ1 ¼ ξdjt þ 1−ξð Þsjt ; ð1Þ

where, ξ and (1 − ξ) are weights assigned to sales and demand,respectively.

Production by firm j is described by a constant elasticity function:

yjt ¼ αK iρKjt þ αL τLiLjt

� �ρ þ αE τEiEjt� �ρ� �1=ρ ð2Þ

where aK, aL and aE stand for the share of capital, labor and electricity inproduction (aKj + aLj + aEj = 1), while iKjt, iLjt, iEjt describe the inputscapital K, labor L and electricity E. In addition, τLj and τEj are labor pro-ductivity and electricity efficiency of firm j, while ρ ¼ σ−1

σ , with σbeing the elasticity of substitution between electricity and capital. Pa-rameters aK, aL, aE, and σ are set equal for all firms. The choice of CESfunction is motivated by the fact that this function is most “rebound-flexible”, in a sense of being capable of accommodating different typesof the rebound effect (Saunders, 2008). In addition, Van den Werf(2008) estimates a two-level CES function with three inputs, and inves-tigateswhether the use of the single-level CES function is appropriate or




not. He finds that the nesting structure, where capital and labor arecombined first, fits the data best. However, he also concludes that formost countries and industries the hypothesis that all three inputs areput into one single nest cannot be rejected. In view of this, we adopt asingle-level CES for analytical convenience.

Labor and electricity productivities can change over time as a re-sult of R&D activities on productivity improvements. The processfollow a three-step procedure: (1) an innovation draw determinethe firm's probability of undertaking R&D activities (0 or 1), follow-ing Nelson and Winter (1982); (2) with the probability proportion-al to relative input prices, a firm directs its research activitiestowards improvements in electricity productivity with the proba-bility pEt

pLtþpEt, or towards improvements in labor productivity with

the probability pLtpLtþpEt

. In the third stage, a firm samples productivity

of an input * (i.e., labor L or electricity E) determined in stage 2 fromthe uniform distribution (0,ν̂�jt), where ν̂�jt is the maximum produc-tivity of input *, i.e. the industry's productivity frontier. The firm im-plements an innovation only if its (randomly drawn) productivity isbetter than its current productivity at time t (ν⁎jt). This means thatnot all innovations are successful. The industry productivity frontierchanges over time (see Eq. (13) later).

Capital is subject to depreciation at a rate δ:

iKjt ¼ 1−δð ÞiKjt−1: ð3Þ

A firm starts with a positive capital at time t=0: iKj0 and it employslabor and electricity so as to produce~yjt+ 1 products. Inputs are allocated

according to their marginal productivity pKt iKjtαK

¼ pLt iLjtαL

¼ pEt iEjtαE

. We substi-

tute iKjt ¼ αKpEtαEpKt

iEjt and iLjt ¼ αLpEt iEjtαEpLt

into ~yjtþ1 ¼ ðαK iρKjtþ αLðτLiLjtÞρ þ αE

ðτEiEjtÞρÞ1=ρ so as to derive the amount of inputs required to produce

~yjt + 1 of products:

iEjt ¼ ~yjtþ1

aEτρe þ aK

aEpKtτρe

pEtaK

� �ρ=ρ−1 þ aLτρL

aEpLtaLpEtτ

ρL

� �ρ=ρ−1 !1=ρ

ð4aÞ

i0Kjt ¼ iEtaEpKtτ

ρe

aKpEt

!1=ρ−1

; ð4bÞ

iLjt ¼ iEtaEpLtτ

ρe

aLpEtτρL

!1=ρ−1

; ð4cÞ

where pKt, pLt, and pEt are prices of capital, labor and electricity attime t, respectively.

A firm invests in capital expansion if iKjt′ N iKjt. In particular, iKjt′ standsfor the amount of capital required to produce ~yjt + 1 products. Thismeans that a firm invests in capital expansion only if the amount of cap-ital which it owns (iKt) is less than capital required to produce the de-sired level of production. The variable cost of production vcjt is:

vcjt ¼ iEjtpEt þ iLjtpLt þ pKtXtn¼t−ι

max i0Kjn−ikjn;0� �

ι: ð5Þ

The last component captures firms' investments in capital during thelast ι time steps. This formula implies that a firm expands its invest-ments costs in period t for ι time periods. This is motivated by the factthat investments in capital are typically considered long-terminvestments.

Firms deposit their profits at the bank. If firm's deposits Djt are lessthan expenses required for expansion, i.e. Djt b vcjt, a firm asks itsbank for a loan ljt = max{vcjt − Djt, 0}. The firm receives the loan to


cover input expenses if its debt-to-equity ratio is less than some critical

ratio (D/E*). The firm's debt-to-equity ratio is computed as ðD=EÞjt ¼∑jt ljt

DjtþpKt iKt−∑jt ljt, where firm's equity (in the denominator) measures

firms' savings (deposits) plus the value of its machinery (capital)minus loans received from the bank.

In case the bank does not have sufficient capital (see Section 3.3), itborrowsmoney in the interbankmarket. The bank contacts other banks,which are drawn with a probability proportional to their liquidity, andissues a loan request. It contacts banks until the liquidity requirementsare satisfied. A firm is obliged to pay back the loan after ι time steps,which represents the average loanduration. Afirm that has not receivedthe requested loan does not expand its production capacity and em-ploys as much energy and labor as it can afford (i.e. total expenses onenergy and labor cannot exceed the amount of its deposits).Wage is de-termined by the supply and demand of labor, as explained inSection 3.2, while the electricity price is determined in the electricitymarket, as described in Section 3.4.

Price-setting by firms follows a simple mark-up rule:

pjt ¼ 1þ ηjt� �

vcjt=yjt þ q xjt� ��

: ð6Þ

Here ηjt is the mark-up, which changes according to past marketshares mjt and parameter ψ as follows (following Dosi et al., 2010,2013):

ηjt ¼ njt−1 1þ ψmjt−1−mjt−2

mjt−2

� �: ð7Þ

Moreover, q(·) captures a fixed cost of producing design x, through amonotonically increasing cost function of the jth design:

q xtþ1ð Þ ¼ γxjt� �ℒ

; ð8Þ

where γ and ℒ are parameters.The quality of the product changes depending on the length of the

period during which the firm produces a particular good zjt, and the

maximum attainable quality x^t at time t (modified based on Malerba

et al., 2001):

xtþ1 ¼ xjt max 0; ~xt−xjt� α1zα2

jt : ð9Þ

Here, the parameter α1 measures the speed of autonomous im-provements towards the maximum attainable quality, and α2 denotesthe competence elasticity. Although α1 is the same for all firms, thepace of improvement in product quality differs among firms. In particu-lar, it depends on how far qualities of firm products are from the maxi-mum attainable quality on themarket. If a firm has no demand, but stillhas savings deposited at its bank, it innovates radically by drawing qual-

ity from (0, x^t). As the process is inherently uncertain the firmmay im-

prove significantly its product quality, but it may also worsen.Firms deposit their profits in the bank at the interest rate r fd. De-

posits change over time, according to:

Dit ¼ Dit−1 1þ r fd� �

þXs¼t

s¼t−τljks−

Xljkt 1þ rlf� �

þ pjtsjt−vjt; ð10Þ

where ∑s¼t

s¼t−τljks are loans from bank k received at t; ∑ ljkt(1 + r lf) are

loan payments made at time t, while pjtsjt−vjt capture net profits(profits minus expenses), with r lf b r fd. In case a firm cannot repay itsloans, it goes bankrupt and leaves the market. There is an exception tothis rule: firms with high markets shares can extend the repayment oftheir loans for the next ι periods. This is motivated by the fact that




some firms, with highmarket shares and profits, are unable to repay theloans taken to cover investments in capital expansion on time.

3.1.1. Entry of new firmsEvery time step, a new firm tries to enter themarketwith some prob-

ability penter. An entrant firm offers a quality ~xt sampled from (0, ~xmaxt),where~xmaxt captures themaximumattainable quality at time t. The latterincreases exogenously over time according to:

~xmaxt ¼ ~xmaxt−1 1þ σqual� �

; ð11Þ

whereσqual captures improvements in the product quality frontier due toeconomy-wide technological innovations.

A new coming firm offers quality that may outperform incumbentproducts. In addition, entrants are likely to adopt more productive tech-nologies than incumbents firms as they are unbound by firm history(including local lock-in) of technology. In particular, labor and electricityproductivities in Eq. (2) are sampled from the uniform distribution (ν�t ,ν̂�), whereν�t is the highest productivity achieved by any firm operatingin themarket at time t, and ν̂�t is themaximumproductivity of input *, i.e.industry's productivity frontier at time t. Themaximumlabor andelectric-ity productivities evolve exogenously, according to:

ν̂et ¼ ν̂et−1 1þ σveð Þ;ν̂lt ¼ ν̂lt−1 1þ σvlð Þ; ð12Þ

where σve and σle capture annual improvements in electricity and laborproductivity, which can be thought of as a result of general progress inthe industry.

The firm j asks a randomly chosen bank k for a loan ljkt to cover initialinvestments in inputs:

ljkt ¼ iLjtpLt þ iFjtpEt þ iKjpKt : ð13Þ

The startup loan intends to cover expenses necessary to produce ygoods. Thus, its size varies with fluctuations in input prices. Inputsequal:

iKjt ¼ yαjt

αK

pKt

pαKKt p

αLLt p

αEEt

ααKK ααL

L ααEE

; iLjt ¼αLpKtpLtαK

iKjt ; iFjt ¼αEpKtpEtαK

iKjt : ð14Þ

We assume that a certain percentage of loans is unsuccessful (1 −psuc), following (Tedeschi et al., 2012). In the first period after enteringthe market, the price of new coming firms pnew is lower than of incum-bents. In particular, a firm is assumed to not impose any markup on thecost of production so as to attract new consumers. This is because a newcoming firm suffers a disadvantage due to the network effect, i.e. con-sumers prefer products with large market shares, which newcomerslack. In addition, incumbent firms offer cheaper products because ofeconomies of scale, resulting in a lower per unit cost of products.

3.2. Consumers

Wedistinguish three types of consumers: nwworkers, nE energy andnK capital owners. Each time period, all firms revise their demand forlabor. We assume a random allocation of workers to firms. Workers re-ceive income from firms. Formally, wages are paid only to nlt ¼ ∑ jiLjtworkers at time t, while nw−nlt of workers remain unemployed. Incase of the excess demand for workers (nw b nlt), wage equals to pLt ¼pLt¼0ð1þ nlt−nw

nwÞ. In this case, all employed workers receive pLt. Other-

wise, workers receive pLt = 0, which can be interpreted as a minimumwage. As a result, wages depends on the demand and supply for labor.Firms' expenses on energy (∑ jiEjtpEt) are distributed equally amongen-ergy owners, while their expenses on capital (∑ jiKjtpKt) are distributed


equally among capital owners. As a result, energy owners receive∑ jiEjtpEt=nE , while capital owners earn ∑ jiKjtpKt=nK .

Each consumer is assigned randomly to a bank at the beginning of asimulation run. The deposits of consumer i (Dit) determines his budgetconstraints. It changes according to:

Dit ¼ Dit−1 1þ rcð Þ þ pLtxwi þ xEiX

jiEjtpEt=nE þ xKi

XjiKjtpKt=nK−pjt ;

ð15Þ

where rc is an interest rate paid on deposits; xwi = 1 if a consumer is aworker and 0 otherwise, xEi = 1 if a consumer is an energy owner and0 otherwise; xKi = 1 if a consumer is a capital owner and 0 otherwise;pjt is the price of the product bought by consumer i.

Each consumer ranks explicitly all products available, and attemptsto purchase the product that yields the highest utility. If the product isout of stock, a consumer attempts to purchase the second-best option.The utility evaluated by each consumer i from adopting a good j de-pends on the product quality xjt, its price pjt (cheapness), the numberof other consumers who bought the product in his/her social class mjt,and the number of workers who purchased it nwjt:

uit ¼ xαijt p

αi−0:5jt mα3

jt n−α4wjt : ð16Þ

The parameter αi captures i's inclination towards the product quali-ty;α3 is the network elasticity; andα4 denotes the snob elasticity (equalto zero if a consumer is a worker). The parameter αi is drawn from theuniform distribution (0, α) for workers, and from (α , α) for energyand capital owners (α b α b 0.5). This assumption implies that energyand capital owners attachmore importance thanworkers to the productquality compared to its price.

In the model, we examine how the presence of two disequilibratingforces, namely a desire for distinction and imitation of other consumerswithin the social network, affects market dynamics. We distinguish be-tween a snob and network effect, which are capturedwith the snob andnetwork elasticity respectively. The snob effect reflects a desire to dis-tinguish oneself from the majority through the purchase of status com-modities. The network elasticity captures the tendency of individuals toimitate choices of others, referred to in the literature as the network ef-fect. Imitation allows saving on costs of individual learning, experimen-tation, or searching by exploiting information already acquired byothers. It may be the source of additional advantages, such as the crea-tion of a network of users, as in telephone and computer markets(Katz and Shapiro, 1985). In themodel, we assume that energy and cap-ital owners (as a more affluent group than workers) try to distinguishthemselves from workers. Safarzynska and van den Bergh (2010a,b)have shown that the snob effect can be an important factor preventingor un-doing lock-in to a single technology.

3.3. Banks

We follow Thurner and Polenda (2013) with respect to modellingthe interbank market. Each time period, banks (1) pay an interest raterc on consumer's deposits, (2) pay an interest rate r fd on firm's deposits,(3) pay an interest rate red on deposits of electricity companies, (4) issuenew loans at the interest rate r lf to new coming firms and at a rate r le toelectricity companies to set up a new power plant, and (5) borrow orlend in the interbank market at the interest rate r IB.

We make a simplifying assumption that all banks offer the same in-terest rate to firms in themarket for final goods. Yet, the interest rate of-fered to electricity companies is lower than in the market of final goodsto reflect the lower risk of investments here. The former is a somewhatunrealistic assumption, as typically interests rate on firms' loans vary soas to reflect the perceived risks of firms. However, if many banks in ourmodel have often one firm as a client, offering differential rates wouldonly affect profitability of banks, increasing the chances of bankruptcies




of banks offering lower inertest rates (see Thurner and Polenda, 2013for the discussion).

The assets of bank k include: reserves Rkt; loans to firms (producers

and power plants)LFkt ¼ ∑ j ∑

t

s¼t−ιljt ; and loans to other banks jLIBkt ¼

∑ j≠k ∑t

s¼t−ιlIBkjt . The liabilities of the bank k are deposits of firms DF

kt ¼

∑ jDjt, and of consumersDCkt ¼ ∑iD

Conit , and loans fromother banksBIB

kt ¼

∑ j≠k ∑t

s¼t−ιlIBjkt . Subsequently, reserves of the bank change according to:

Rkt ¼ Rkt−1 þ 1þ rIB� �X

jlIBkjt−τ−

XjlIBkjt þ Dc

kt−Dckt−1 þ

XjlIBjkt

−ð1þ rIBÞX

jlIBjkt−τ ;

ð17Þ

where ð1þ rIBÞ∑ jlIBkjt−ν are loans repaid by other banks;∑ jl

IBkjt are loans

given at time t to other banks; ∑ jlIBjkt are loans received; and ð1þ rIBÞ

∑ jlIBjkt−ι are loans repaid by bank k at time t.The loan received by bank k from bank j, depends on the sum re-

quested by bank k lk_reqIB and liquidity of bank j. In particular, bank j bor-

rows to bank k:

lIBjkt ¼ lIBk req if Rjt−lIBk req ≥ Rjtmin;

lIBjkt ¼ Rjt−Rjtmin otherwise:ð18Þ

Rjtmin are theminimum reserves required by central banks to be heldby commercial banks to prevent bankruptcies because of unexpectedlosses. The minimum reserves are computed as a fraction of deposits:

Rjtmin ¼ max β DCjt þ DF

jt

� �;R

� �; ð19Þ

where R is the minimum level of reserves regardless of deposits.The equity of a bank is given by subtracting its liabilities from assets:

EqFit ¼ Rkt þ LIBkt þ LF

kt þ LEkt−DEkt−DE

kt−Dckt−BIB

kt : ð20Þ

The equity of a bank is not affected by changes in reserves, but onlyby interest payments or credit defaults of firms and banks. A bank goesbankrupt if either its equity or its reserves become negative. There is norecovery of interbank loans considered, and lending banks write off theloans to the default banks.

We introduce additional assumptions to Thurner and Polenda(2013). In particular, in case a firm goes bankrupt, its capital stock aug-ments the stock of capital owned by the bank. The bank resells it to newcoming firms with the probability presell. This allows banks to recoversome losses in case of firms' bankruptcies. In addition, in our model,banks use the debt-to-equity of firms as an indicator of riskiness offirms' investments. Banks can also extend repayment of loans to firmsthat have sufficiently large market shares (N0.2 of the market) and arecharacterized by a positive equity. In this way we prevent bankruptingof firms that have not accumulated sufficient profits to repay theirdebts on time, but are likely to repay these debts in the future.

Finally, we do notmodel explicitly the entrance of new banks,whichis of course an unrealistic assumption. Yet, it helps us to visualize bank-ruptcies of banks and to examine the possible effects of the cascades offailures in the banking sector on the real economy. An alternative as-sumption, which can be encountered in the literature, is to replacebanks, which go bankrupt, with the entry of new banks. However,such new comers would be prone to bankruptcies unless assigned asubstantial amount of reserves initially. This would cause the monetarybase to change over time beyond our control (we keep the monetarybase constant over time).


3.4. Electricity market

The electricity market is modeled based on Safarzynska and van denBergh (2011) and Safarzynska (2012). On the electricity market, pro-duction of electricity is carried out in three heterogeneous plants i char-acterized by age sit, specific productivityνit and energy source j, installedcapacity ki, maximum lifespan Tj, and capacity factor λj. The latter cap-tures periods of decreased production due to economic reasons (lowprofitability), obligatorymaintenance, etc.We distinguish three stylizedfuels competing for adoption: fuel 1 is characterized by the unit cost in-creasing over time; fuel 2 is described by the units cost decreasing overtime (according to the Brownian motion equation below); and fuel 3 ischaracterized by constant price of fuels.

The structure of dynamics on the electricity market is as follows. Atthe beginning of each time step t, plants set their production qit so asto maximize profits:

Πit ¼ pEt 1þ ηp� �

qit−mitqit ð21Þ

where pet is the spot market price of electricity, ηp is the markup im-posed on the electricity price, and mit is a marginal cost of plant i. Theelectricity price is determined by an inverse demand function:

pEt ¼ a−btQt þ θ ð22Þ

where Qt is equal to a total supply:Qt ¼ ∑iqit, and a and bt are param-

eters. θ is a random variable drawn from normal distribution N(0,1). Toensure that electricity produced in the electricity market is equal elec-tricity demanded in themarket forfinal goods, the parameter bt changes(derived based on Eq. (4a), by equating the inverse demand for electric-ity in themarket for final goodswith Eq. (22), and substituting the totalsupply Qt ¼ ∑

iqit) as:

bt ¼ neaþ neθ−Mtð Þne þ 1ð Þ yy−ρ o

aþMt þ θnE þ 1

� �−ρ=ρ−1

þ aEτρE

! !1=ρ

; ð23Þ

where o ¼ aK ðaEpKtτρe

aKÞρ=ρ−1 þ aLτ

ρL ðaEpLtaLτ

qLÞρ=ρ−1 and yy is the expected total

sales at themarket for final goods. In particular, we assume that electric-ity producers expect firms to produce the same amount of products asin the past period.

In Eq. (21), we introduced a markup on the electricity price. This ismotivated by the fact that electricity demand is very volatile, amongothers, because of cyclical demand for final goods, the entry and exitof new power plants and firms in the market for consumer products.As a result, past demand is only an approximation of the future demand.In initial simulations with the markup equal 0, demand for electricityvery often exceeded supply. This is an unrealistic scenario, underwhich some firms would be unable to buy inputs for production. Toavoid such situations, we introduced a markup, which drives the elec-tricity price up. The markup causes that supply always slightly exceedsdemand for electricity, while the price is sufficiently high to minimizepotential losses due to unsold electricity.

The production decision by electricity plants ismodeled as a Cournotgame. Accordingly, each plant decides how much output to produce so

as to maximize profits (derived from ∂πit∂qit

¼ 0):

qit ¼aþ θ− ne þ 1ð ÞmitþMt

ne þ 1ð Þbt : ð24Þ

Here, ne is the number of power plants operating at the electricitymarket, andMt is the sum of marginal costs of all power plants operat-ing at time t.




After setting production, plants decide howmany inputs for produc-tion to employ so as tominimize total input costs. Electricity productionby plant i using technology j is described by the Cobb-Douglas function:

qit ¼ ai iEKit� �αE

Kj iELit� �αE

Lj iEFit� �αE

Fj; ð25Þ

where ait is the plant's specific productivity; iKitE , iLitE , and iFitE describe

capital, labor and fuel input respectively. αKjE , αLj

E , and αFjE are corre-

sponding substitution factors associated with technology j, where

αKjE + αLj

E + αFjE = 1. The parameter ai is equal to ð1viÞ

αEFj , where vit is

a thermal efficiency with which a plant can transform fuel intoheat (energy).

Under the assumption that inputs are allocated according to theirmarginal productivity, inputs are equal:

iEKit ¼αEKitpFjt

pojtαEFjiEFitvi; i

ELit ¼

αELjpFjt

pLtαEFj

iEFitvi and iEFit

¼ qitαEFJ

pFjt

poj� �αE

Kj pαELj

lt pαEFj

Fjt

vi αEkj

� �αEkj αE

lj

� �αElj αE

Fj

� �αEFj

; ð26Þ

where pL is wage, pFjt are the prices of fuel j at time t respectively and pjo

is the operating costs. We assume that the price of labor is equal towages in the market for final products. The operating cost equals tothe cost of capital, but it varies among different energy technologies soas to reflect their different capital intensities (or different initial capitalrequirements).

Prices of fuels change over time. In particular, fuel prices follow ageometric Brownian motion (Brandt and Kinlay, 2008):

dpFjt=pFjt ¼ χ jdt þ σ jdZt ; ð27Þ

where σ is the volatility of fuel price j, Zt is a Wiener process and χ is adrift.

The totalcostofproduction isdefinedasTCit= iKitE pjt

o+ iLitE pLt+ iFit

E pFjt;Fjt; from this themarginal cost of the power plant i employing technol-

ogy j can be derived as being equal to (ðmit ¼ ∂TCit

.∂qit

Þ:

mit ¼viαE

Lj þ vitαEKj þ αE

Fj

� �αEFj

� �αEKjþαE

Lj pojt þ pLt� �αE

KjþαELj

viαEFj αE

Kjpojt

� �αKjαELjpLt

� �αELj

ð28Þ

where poj is the operating cost of technology j.A plant exits once sit N Tj where Tj is the expected lifetime of a plant

(defined for each energy technology). Afterwards, the owner invests ina new power plant. Formally, an owner (1) evaluates capacity kij re-quired to produce qit electricity given marginal costs of different energytechnologies j and the total marginal costs of incumbent plants(Eq. (22)), and (2) compares expected profits Vij for each energy tech-nology j:

Vij ¼XTþ1

t¼1

e−rEt p λijkij� �

−m̂jt� �

λijkij−e−rE I jkij

¼ −1−1þ erE

ðe−rE Tþ2ð Þ −1þ erEþTrE� �

kijðI j þ erEλij c−p kij� ��

: ð29Þ

Here, Ij is a fixed cost per KWof installed capacity kij capturing initialinvestment costs andmaintenance expenses, and rE is the discount rate.These costs need to be covered from the revenues over the entire life ofthe plant Tj. Furthermore, m̂jt is the expected marginal cost associatedwith technology j at time t + 1. A new plant embodies technology jthat ensures the highest value Vij. According to Eq. (29), an owner ofthe future power plant calculates how the current electricity price


would be affected by the installation of a new power plant (p(λijkij)),and then estimates discounted value of investments in such a plant. Anew power plant that enters the market receives a loan from the bankto cover costs of investments in installed capacity (Ijkij). It is obliged topay it back when it reaches it maximum lifespan.

4. Results frommodel simulations

We discuss empirical validation of our model in Section 4.1. InSection 4.2, we study robustness of our results to changes in key param-eter values so as to better grasp mechanisms underlying modeldynamics.

4.1. Results from the baseline simulations

Each simulation run lasts for 1000 time steps, where a single stepcan be interpreted as a year. Unless stated otherwise, we repeated sim-ulations 100 times for each version of themodel (with different param-eter settings; see Table 1), in order to check the stability of our results inthe presence of stochastic factors (i.e. a Monte Carlo analysis). In thissection, we discuss illustrative model dynamics in the baseline scenarioandmotivate the choice of our parameters. In the next section,we studysystematically the robustness of model outcomes to changes in key pa-rameters, which we identified as crucial for model dynamics in initialsimulations with many parameters randomly generated.

The baseline parameter values are described in Appendix A. Theseparameter valueswere chosen so as the baseline scenario generates sta-ble labor and electricity markets, as well as the banking sector. For in-stance, altering initial values of labor and energy productivities doesnot affect model dynamics, but only the mean level of employmentand electricity used in our simulations. Therefore, we chose parametervalues that generate full employment. This allows us to study conditionsunder which the economy goes into the recession (i.e. employmentsdrops suddenly and significantly below full employment). The parame-ter values describing the electricity market were calibrated based on1990–2002 data from the British electricity industry, followingSafarzynska and van den Bergh (2011) and Safarzynska (2012). Howev-er, we had to re-scale those parameters so as dynamics in the electricitymarket in the currentmodelmatch demand for electricity in themarketfor final goods. For instance, we find that the installation costs of alter-native energy technologies determine the type of fuel embodied bynew power plants and their size. In the baseline scenario, we assumethat the fuel characterized by a constant cost is the cheapest to installso as to isolate the effect of improvements in energy productivitiesfrom effects of fluctuations in fuel costs on model dynamics. We relaxthis assumption in Section 5.

We show thatmodel simulationswith the baselineparameter valuesare capable of replicating awide spectrumof stylized facts. In employingthis method of validation, we follow the approach used in Dosi et al.(2013, 2015). In particular, we find that employment and total debt inthe economy are procyclical (Stock and Watson, 1999; Dosi et al.,2015), while loan losses (as a percentage of total loans) in the economyare countercyclical (Bikker and Metzemakers, 2005). In testing the cy-clical behavior of macroeconomic variables we follow the methodologyby Summer and Silver (1989). In particular, we regress the first differ-ences of the logs of variables in questions on time and the first differ-ences of the log of total capital using data generated by oursimulations (see results in the Appendix B, Table B1).

Fig. 2 presents illustrative results from a typical simulation run for:(a) electricity use and employment, (b) capital growth; (c) the totalnumber of products sold (figure at the top) and production by differentfirms (figure at the bottom); (d) the number of banks surviving overtime (the figure at the top), and the corresponding connectivity in theinterbank lending market (the figure at the bottom). The connectivityis measured as the number of linkages between banks (per bank) in



Table 1Mean statistics under different scenarios.

The mean values of: Scenarios

Baseline (both productivities;strong network)

Weak snob andnetwork effects

No improvementsin productivities

Low productivities Only improvements inenergy productivity

A higher interest rate(equal to 0.1%) on consumerand firms deposits

Growth of capitala 0.006 0.008 0.008 0.007 0.004 0.003(0.04) (0.01) (0.06) (0.08) (0.06) (0.08)

Growth of capitalb 0.04 0.06 0.04 0.04 0.04 0.04(0.02) (0.01) (0.02) (0.02) (0.01) (0.02)

The number of products sold 364.41 440.36 357.93 352.38 371.69 367.16(27.97) (18.46) (26.13) (28.08) (27.63) 32.22

Workers and the correspondingmean unemployment rate

489.06 478.25 739.63 762.47 791.54 491.57(21.05) (11.59) (19.25) (34.33) (21.45) (24.45)

0.25 0.26 0 0 0 0.24Energy consumption 355.99 346.94 539.75 555.18 425.05 356.95

(13.68) (7.96) (14.11) (20.90) (14.73) (15.86)The number of banks in thelast period

39.87 21.02 40.21 39.66 42.92 26.12(3.98) (4.27) (5.55) (3.45) (5.53) (11.12)413.15 406.46 413.42 419.32 406.87 415.18(19.60) (6.73) (12.48) (19.32) (13.36) (27.16)137.51 140.34 137.29 110.63 136.05 144.74(34.58) (52.56) (70.74) (28.06) (32.76) (31.24)229.03 227.74 297.81 236.31 315.35 228.35(107.52) (39.97) (204.59) (189.37) (238.57) (79.23)368.3 291.91 494.77 433.55 590.93 341.02(190.84) (47.71) (280.69) (274.37) (312.84) (146.81)

Note: Standard deviations in brackets.a Growth of capital is computed as (totalcapitalt − total capitalt − 1) / total capitalt − 1.b Growth of capital is computed as |totalcapitalt / total capitalt − 1 − 1.


the interbank lending market. Finally, Fig. 2(e) shows how total debt inthe economy changes over time.

Fig. 2(a) illustrates that employment as well as electricity used inmanufacturing are declining because of the improvements in inputs'productivities over time. The mean values of electricity and labor in100 simulations of the baseline scenario (over the entire simulationrun) are equal to, respectively: 356 energy units (sd. 13.68) and 489workers (sd. 21.50). The average growth of capital in 100 simulationsequals 0.6% (±1.7%), which is not unusual for developed economies(Fig. 2(b)). For instance, in 2013 its growth rate for Austria, France, orGermany was below 0.5% (World Bank, 2014). In Fig. 2(b), substantialdeclines in the capital stock are caused by a sudden bankruptcy of adominant incumbentfirm.When a new firm enters themarket, produc-tion gradually expands: at first it receives a startup loan to produce 30products. Only if its product is demanded by consumers, a firm receivesa loan, so that it can expand its capital. The spikes in growth of capitalcorrespond to periods of capital expansion by new firms.

Wefind that ourmodel is able to robustly generate sustained growthcharacterized by business cycle (Fig. 2(d)). The persistence of businesscycles is closely linked to entry and exit of new firms as well as a cyclicaldemand. In particular, workers cannot afford purchasing products eachtime step and have to save money. As a result, they buy products everyfew time steps. This explains the resulting cyclical patterns of demandand employment. Until time step 200, the market is dominated byfirm 2 (Fig. 2(e)), which is due to the strong network effect assumedin the baseline. Around time step 200 the economy goes into a recessionas the main incumbent firm (firm 2 in Fig. 2(e)) goes bankrupt. This isbecause themaximum product quality increases over time due to exog-enous technological progress. As a result, a new firm (firm 4) enters themarketwhich offers higher quality than incumbents. The quality offeredby new firmsmay offset the network advantage of the incumbent firms.Afterwards, we observe a lower labor and electricity use (Fig. 2(a)),which can be explained by the fact that entrant firms employmore pro-ductive technologies. As a result, they use less labor, which further un-dermines demand.


Fig. 2(c) shows the evolution of surviving banks in themarket andthe corresponding connectivity of banks in the interbank lendingmarket. On average, 40 (±4) banks survived in the last period inthe baseline scenario. We will refer further in the text to the numberof surviving banks as it is indicative of the severity of cascades of fail-ures in the banking sector. We do not assume this to be a measure ofgeneral stability of the financial sector, as a banking sector with onlyone bank can be stable. The figure shows that banks' collapses arepreceded by an increasing connectivity of banks in the interbanklending market, as well as an increasing debt in the total economy(Fig. 2(e)). In particular, banks' failures around time step 200 werepreceded by a significant increase in the total debt in the economy.This was caused by incumbent firms taking loans to expand theirproduction capacities, as well as due to substantial loans given topower plants in the electricity market. The loans in the electricitymarket decrease over time. As the productivity of electricity im-proves, firms use less electricity in production, and thus smallerpower plants are installed over time.

4.2. Robustness of model outcomes to key parameter values

In this section, we discuss sensitivity of model results to changes inkey parameters. Section 4.2.1 studies how the network and snob effectsaffect economic and financial stability. Section 4.2.2 discusses the im-pact of changes in electricity and labor productivities on model out-comes. Section 4.2.3 presents results from the panel regressions withmany different parameters randomly generated so as to identify factorsconductive to financial and economic stability.

4.2.1. Network and snob effectsThe current economic system relies on unlimited growth of material

production and consumption. After satiation of existing wants, peopleactively seek newmeans to satisfy their desires, which creates a viscouscycle of consumption. The search of new consumption possibilities isoften driven by status competition (Frank, 2008). However, there is



(a) Labor/electricity use (b) Capital growth

(c) Banks’ connectivity and nr. of banks (d) Total number of products sold

(e) Total debt

Fig. 2. Illustrative model simulations. Note: Growth of capital in (b) is computed as (total capitalt − total capitalt − 1) / total capitalt − 1.


little research on how important is status competition for macroeco-nomic outcomes and environmental goals. With our research we aimat filling in this gap. In particular, we compare model outcomes in thepresence of the strong and weak network and snob effects. Fig. 3ashows the results for a typical run in the presence of the strong networkand snob effects (α3=0.6,α4=0.5),while Fig. 3b does the same for theweak network and snob elasticities (α3 = 0.01, α4 = 0.3).

Our illustrative figures indicate there is a tension between theneed for distinction and conformity between members of differentclasses that causes the clustering of consumer choices around twodifferent products. In Fig. 3a in the presence of the strong networkeffect, a single firm dominates the market during first 500 steps.On the other hand, Fig. 3b shows that within the 250 first timesteps many products compete in the market and sequentially


dominate the market under the weak network effect, resemblingthe fashion market. New products compete for adoption and can at-tract consumers even if they have not captured large market shares.This is because brand loyalty is not important here. In addition,commodities that initially confer social status become less attrac-tive when more workers start adopting them. Subsequently, moreaffluent consumers shift to products that can distinguish themfromworkers. As a result, cyclical patterns or waves of consumptionemerge. Similarly, Janssen and Jager (2002) show that if consumersengage in social comparison and deliberation, market dynamics re-semble a ‘fashion market’, while in the model developed by Saviottiand Pyka (2008) cyclical patterns of industry life cycles emergefrom the interplay of technological competition and marketsaturation.



(a)Baseline values of parameters (b) Weak network and snob effects

(c) The number of surviving banks (d) The number of surviving banks(randomly generated the network elasticity) (randomly generated the snob elasticity)

(e) The mean number of the log of workers (f) The mean number of the log of workers(randomly generated the network elasticity) (randomly generated the snob elasticity)

Fig. 3.Model dynamics in the presence of strong (network elasticity = 0.5; snob elasticity = 0.6) and weak network externalities (network elasticity = 0.01; snob elasticity = 0.3).


We compare mean results from 100 simulations between twoscenarios. In the first scenario, we use baseline parameters, andthus the network and snob externalities are strong. In the secondscenario, the network and snob externalities are weak. There areno significant differences in the mean of workers and of electricityuse over time between two scenarios. However, significantly morebanks survived by the end of the simulation period in the presenceof the strong network effect (39 ± 4) compared to the weak net-work effect (21 ± 4). This is related to the fact that the mean con-nectivity of banks in the interbank lending market was higher inthe latter case (0.78 ± 0.19) compared to the baseline (0.66 ±0.25). In particular, the more firms compete for adoption, the moreloans are requested by firms who want to expand their productioncapacities. In the presence of the strong network effect, banks assessthe riskiness of projects based on firms' debt-to-equity ratio andtheir markets shares, before granting loans. However, these indica-tors turn out to be less indicative of future potential of firms torepay their loans under the weak network effect, where consumers


easily change brands. As a result, more loans are grants to firms,which are then unable to pay them back, causing a higher rate offirms' and banks' bankruptcies.

To examine how network and snob elasticities influence financialand economic stability, we conducted 100 simulations with the net-work effect randomly generated from (0,1) and other parameter valuesas in the baseline; and an additional 100 simulations with the snob ef-fect randomly generated from (0,1). Fig. 3(c) and (d) depict the numberof surviving banks in the last period depending on the value of the net-work effect (in case the network effect was randomly generated) anddepending on the value of the snob effect (in case the snob effect wasrandomly generated), respectively. Fig. 3(e) and (f) do the same forthemeanvalue ofworkers over the simulation period. The depicted pat-terns reveal that the higher the values of snob and network elasticities,the more banks survive in the last period up to critical values of theseparameters (i.e. around 0.6 of the network elasticity and 0.7 of thesnob elasticity). Increasing values of these parameters further has a neg-ative impact on the number of surviving banks. On the other hand,



Table 2Results from the logit panel regression with the dependent variable taking value 1 if a sin-gle bank collapsed in time t in Column 2; from the random-effects model with the depen-dent variable being the logarithm of employment in Column 3.a

Dependent variable: Banks' collapse Random-effects modelLog of employment

Growth of capitalt − 1 −0.28 −0.01⁎⁎⁎

(0.31) (0.001)Growth of energyt − 1 −0.37⁎ 0.003⁎⁎⁎

(0.16) (0.001)Banks' connectivity at the 0.82⁎⁎⁎ 0.03⁎⁎⁎


increasing the value of the network elasticity has atfirst negative impacton the mean value of the logarithm of employment, until the networkelasticity reaches around 0.5. Afterwards, an increase in the networkelasticity has a negative impact on the mean employment. Similarly,the impact of the snob effect on employment is also nonlinear. Resultsfrom panel regressions in Table 2 support these results. The snob andnetwork elasticity turn out to have significant and negative impact onthe probability of banks' collapse as well as on the logarithm of employ-ment, while the squared values of these elasticities have significant andpositive impact on the dependent variables.

interbank lending markett − 1 (0.02) (0.001)Network elasticity −2.29⁎⁎⁎ −0.45⁎⁎⁎

(0.72) (0.08)Snob elasticity −3.79⁎⁎⁎ −0.25⁎⁎⁎

(0.65) (0.07)A squared value of network elasticity 2.27⁎⁎⁎ 0.44⁎⁎⁎

(0.69) (0.03)A squared value of snob elasticity 3.04⁎⁎⁎ 0.14⁎⁎

(0.68) (0.07)Constant −15.06⁎⁎⁎ 7.00⁎⁎⁎

(0.55) (0.07)Nr. of observations 199,360 199,360Nr. of groups 200 200Wald Chi2 1790.62 798.42R2 withinBetweenOverall

0.0030.330.01

Note: standard deviation in parenthesis.⁎⁎⁎ Indicate variables significant at the 1% level.⁎⁎ Indicate variables significant at the 5% level.⁎ Indicate variables significant at the 10% level.a We included as independent variables: past growthof capital and energyuse; past values

of banks' connectivity in the interbank lendingmarket, and values of snob and network elas-ticities, as well their squared values. The sample includes 200 observations from the baselinesimulation: in 100 cases the network effectwas randomly generated from (0,1); in 100 othercases the snob effect was randomly generated from (0,1).

4.2.2. Energy and labor productivitiesThe decoupling of environmental pressures from economic growth

has been suggested as a solution to environmental problems. Yet, globaleconomic growth and energy intensity have both contributed to in-creasing energy use in recent years, creating skepticism about thefeasibility of decoupling. In this section, we examine the effects of im-provements in electricity and labor productivities on model outcomes.

In this section,we compare economic andfinancial stability in highlyproductive economies (Scenario 1, denoted “both productivities” inFig. 4); in moderately productive economies (Scenario 4, denoted“low” in the figure)3; and economies, where no investments are madein labor and energy productivities (Scenario 2, denoted “no improve-ments”), or only investments in productivities of electricity are under-taken (Scenario 3). So far, existing macroeconomic models typicallyignore firms' innovative activities in energy productivity, while to ourknowledge no other model allows firms to choose between investingin labor or energy productivities depending on the relative prices of in-puts. Below, we explore the consequences of these simplifying assump-tions in conventional models. For each productivity scenario, we run100 simulations in the presence of weak (ς=0.01, κ=0.3) and strong(ς=0.6, κ=0.5) network and snob effects (mean results are reportedin the Appendix in Table B2).

Fig. 4(a) compares the number of banks surviving in the last periodbetween different scenarios for strong and weak network effects sepa-rately. We test for statistically significance differences in the numberof surviving banks using the Wilcoxon-Mann-Whitney test betweendifferent productivity scenarios in the presence of the strong networkeffect.Wefind that statistically significantlymore banks survives in Sce-nario 3 compared to each of the other scenarios (p b 0.001). There are nostatistically significant differences in the number of banks betweenother scenarios (1, 2 and 4) in the presence of the strong network effect.This can be explained by the fact that labor constitutes amore expensivefactor of production in our model causing firms to invest in improve-ments in labor productivity more often than in electricity productivity.In turn, the more productive labor is, the fewer workers are employedin the economy. As a result of high unemployment, demand is lowand firms are less likely to pay back their loans, which causes a higherrate of bank bankruptcies.

In the presence of theweak network effect, significantly fewer bankssurvive by the end of simulations, which confirms results from the pre-vious section. In addition, there are statistically significant differences inthe number of surviving banks between each of the energy scenarioshere. Our results suggest that if brand loyalty is unimportant, the sever-ity of cascades increases the less innovative economy is.

Fig. 4(b) shows for each productivity scenario the mean values ofemployment and electricity use over 100 simulations.We show only re-sults for the strong network and snob effects as the distributions of em-ployment and electricity use are quite similar under weak and strongnetwork effects (see Table B2). We find that more inputs are used in

3 Here, firms can still invest in both labor and electricity productivities, but the maxi-mum attainable productivities are half of their baseline values.


the absence of productivity improvements than in the baseline scenario,while fewer inputs are used in the baseline scenario compared to thelow productivity scenario, in line with theoretical expectations. Con-trary to our expectations, electricity use turns out to be much higherunder the ‘only improvements in electricity productivity’ scenario com-pared to the baseline scenario. This suggests the presence of a so-calledenergy rebound effect, that is, indirect additional energy use due to im-proved energy efficiency or conservation. Two important mechanismsto explain this are more intense use of efficient technologies as theyhave become effectively cheaper, and re-spending of monetary savingsassociated with energy savings (Sorrell, 2007; Antal and van den Bergh,2014). In our model, if firms invest only in improvements in electricityproductivity, the mean employment is higher compared to the baselinewhere firms invest in improvements in both productivities. As a result,consumers can buy more products, whose production requires electric-ity. Total electricity use can thus come out higher than under the base-line scenario.

4.2.3. Factors conductive to financial and economic (in)stabilityIn this section, we present results from statistical analysis (panel re-

gressions) of the data generated by our model simulations so as to un-derstand factors conductive to the cascades of bank failures andunemployment. We follow the approach of Windrum and Birchenhall(2005), who estimate a probability of technological succession in thelogit model using data generated by their simulations.

To check the robustness of our results to changes in key parameters,we run 150 simulations with different parameter values (generatedrandomly before each simulation). Table 3 summarizes results fromthe panel regressions. In Column 2, we present results from the logit re-gression with the dependent variable taking a value of 1 if any bank



(a)Number of banks surviving in the last period (b) Mean electricity consumption and the number of workers under strong network and snob effects

Fig. 4.Model outcomes for different productivity scenarios.


went bankrupt at time t and 0 otherwise. Column 3 of the table presentsthe results from the random-effects model, where the dependent vari-able is: a logarithm of total employment at time t, and Column 4presents results from the fixed-effects model. In both cases, we addedAR(1) disturbances into regressions to correct for serial correlation asthe Wooldridge test indicated that the data suffer from autocorrelation(F-statistic = 342.96). The choice of a random effects model was

Table 3Results from the logit panel regressionwith the dependent variable taking value 1 if a sin-gle bank collapsed in time t in Column 2; from the random-effects model with the depen-dent variable as the logarithm of employment in Column 3; and from the fixed-effectsmodel with the dependent variable equal to the logarithm of employment in Column 4.

Dependent variable: Banks collapse Random-effectsmodel Log ofemployment

Fixed-effectsmodel Log ofemployment

Maximum labor productivity 0.88⁎⁎⁎ −0.18⁎⁎⁎ −0.21⁎⁎⁎

(0.07) (0.01) (0.01)Maximum energy productivity 0.87⁎⁎⁎ 0.07⁎⁎⁎ 0.06⁎⁎⁎

(0.07) (0.01) (0.01)H-indext − 1 0.04⁎⁎⁎ 0.03⁎⁎⁎

(0.01) (0.004)Frequency of new firmsentering the market

0.16 −0.14⁎⁎ −0.14(0.69) (0.10) (0.10)

(D/E)* 0.006⁎⁎⁎ 0.002⁎⁎⁎

(0.002) (0.001)A frequency of energy owners 7.54⁎⁎⁎ −1.02⁎⁎⁎

(0.75) (0.10)A frequency of capital owners 4.95⁎⁎⁎ −1.16⁎⁎⁎

(0.76) (0.12)Banks' connectivity at theinterbank

lending markett − 1

0.58⁎⁎⁎ 0.06⁎⁎⁎ 0.05⁎⁎⁎

(0.03) (0.003) (0.003)

Network elasticity −0.54⁎

(0.29)0.09⁎⁎⁎

(0.04)Variance of debtt − 1 3.71⁎⁎⁎

(0.12)Constant −15.06⁎⁎⁎ 7.00⁎⁎⁎ 6.36⁎⁎⁎

(0.55) (0.07) (0.01)Nr. of observationsNr. of groups

149,700 149,691 149,541150 150 150

Wald Chi2 1580.68 1747.02 414.65R2 withinBetweenOverall

0.03 0.010.69 0.180.12 0.04

Note: standard deviation in parenthesis.⁎⁎⁎ Indicate variables significant at the 1% level.⁎⁎ Indicate variables significant at the 5% level.⁎ Indicate variables significant at the 10% level.


motivated by the Breush-Pagan test, which indicated that this was pref-erable over OLS regressions. The Hausman test indicates that fixed-effects model is preferable over random-effects model (Chi2 =830.53). However, afixed-effectsmodel does not allowstudying the im-pact of time invariant variables, which is the aim of this section. There-fore, we provide results from the two models. Table 3 shows that bothprovide similar results.

We included as dependent variables parameters randomly generat-ed before each simulation run, such as: themaximum labor and electric-ity productivities (drawn randomly from U(1,2)); the frequency of newfirms entering themarket (drawn randomly from U(0.01,0.2)); the net-work elasticity (drawn randomly fromU(0.01,0.6),while the snob effectwas set to 0.3); the debt-to-capital ratio above which banks refuse togrant loans to firms (from U(5,50)); the frequency of capital and energyowners in the population (both drawn randomly fromU(0.05–0.25)). Inaddition,we added as independent variables to the regressions: the pastvalues of banks' connectivity in the interbank lending market; the pastvalue of growth in variance of banks' loans (the variance of debt inTable 3); and the Herfindahl index. The latter is computed as: ∑ jm

2jt

is mjt are the market share of firm j in time t. The connectivity measureis defined as the average number of connections between banks in theinterbank lending market.

The results from the panel regressions in the second and third col-umns confirm previous results that the stronger the network effect is,the less likely banks go bankrupt. On the hand, the more frequentlyfirms enter the market, the more likely the cascades of banks' failuresoccur. However, the frequency of new firms is an insignificant predictorof the probability of banks' collapses, unless the network effect is weak.Adding the interaction term between the frequency of entry of newfirms and a dummy taking a value 1 if the network effect is below 0.1(and 0 otherwise) turned out to be a significant predictor of the proba-bility of banks' collapse (we do not report the resultswith the additionaldummy). This can be explained by the fact that in case brand loyalty isnot important, model dynamics resemble the fashion markets(Fig. 5a). As a result, loans to firms, which temporarily dominate themarket, are more likely not to be paid back, rendering the cascades ofbanks' failures. This effect is stronger, the higher the probability ofnew firms entering themarket is (Fig. 5b). In addition, during the periodof new firms expanding their production capacities, periods of high un-employment may occur (Fig. 5c). Results in the third column of Table 3confirm that the stronger network effect generatesmore stable employ-ment. This result is confirmed by the positive and significant impact ofthe Herfindahl index on employment. Note that a higher value of thisindex mean less competition, while a value of 1 indicates the marketis a monopoly.



(a) Firms' production (frequency of new firms=0.3) (b) The number of surviving banks

(c) Energy use and employment (frequency of new firms =0.3)

Fig. 5.Model dynamics in the presence of the weak network effect (network effect = 0.01, snob elasticity = 0.3).


Other results in Table 3 indicate that bankruptcies of banks are pre-ceded by a relatively high connectivity in the interbank lendingmarket,which is turn is driven by growth in total debt, i.e. total loans issued bybanks. In the theoretical literature, there is no consensus on the effect ofincreasing network connectivity on the risk of banks' bankruptcies. Onthe one hand, an important class of models, namely where the creditsystem is depicted as a random graph, have shown that increasing net-work connectivity decreases the probability of banks' avalanches (Allenand Gale, 2000; Thurner and Polenda, 2013; Iori et al., 2006) because ofrisk sharing. On the other hand, increasing the connectivity may causethe network of banks to be less exposed to systemic risk only in anearly phase, while too high connectivity (above a certain threshold)may increase systemic risk (Iori et al., 2006; Lorenz et al., 2009).

In addition, we find that in more productive economies, i.e. charac-terized by higher labor and electricity productivities, the probability ofbanks' failures is higher. This can be explained by the fact, that themore productive inputs are, in particular labor, the less income as ashare of lower firm's profits comes back to workers as wages. This di-minishes workers' purchasing power. Finally, we find that higher fre-quencies of capital and energy owners decrease the probability ofbanks' collapses. The higher frequencies of these groups imply incometo bemore evenly distributed in the economy.Weexamine the latter re-sult more closely in the next section.

5. Policies for macroeconomic, financial andenvironmental sustainability

From a policy angle, a formal modeling approach is needed that cap-tures essential technological and behavioral features of energy and fi-nancial sectors as well as accounts for interactions between associatedsubsystems of the economy. This is because solutions to problems inspecific subsystems, such as energy taxes, labor or financial regulations,may have profound impacts on other subsystems of the economy. Find-ing the right combination of policies requires an analytical tool that


integrates these subsystems to account for their feedbacks. For instance,it has been suggested thatmoney creation by commercial loans can ren-der an increase in the price of energy, destabilizing the economy(Douthwaite, 2012), or that energy taxes may cause the economy togo into debts without inducing necessary changes in consumption pat-terns (Rifkin, 2011). Another proposition is that subsidies for renewableenergy might trigger oil market responses that will affect the rest of theeconomy, rendering more rather than fewer CO2 emissions, a phenom-enon which is currently studied under the label of “green paradox”(Sinn, 2012; van der Ploeg and Withagen, 2010). In addition, shiftingtaxes from labor to energy or carbon is likely to reduce labor productiv-ity innovations, which would temper income growth (Jackson andVictor, 2011; Antal and van den Bergh, 2013). All these examples illus-trate that the way macroeconomics currently works is bound to be in-complete and biased. We need to strive towards an encompassingsystems perspective on the economy if we are genuinely trying tooffer policy advice for a sustainable and stable economy with (closeto) full employment. In this section,we study the impacts of various pol-icies intended to achieve three goals, namely macroeconomic, financial,and environmental sustainability. These scenarios include: an equal in-come distribution; different energy scenarios (energy subsidies); andlending regulations of banks. Evidently, our analysis does not aim tobe exhaustive. It deals with a new topic and merely offers a startingpoint for further research.

5.1. Distributive policy

Is income inequality harmful for growth? Over the last decades, alarge body of theoretical and empirical studies has addressed thisquestion. On the one hand, greater inequality may be detrimentalto growth by inducing distributional conflicts. According to thisview, the greater inequality of wealth and income leads to the higherrate of taxation, and subsequently to lower growth (Persson andTabellini, 1991; Alesina and Rodrik, 1994). On the other hand, high




inequality has been argued to motivate people to work harder, fosteraggregate savings and capital accumulation by the rich as they have alower propensity to consume (Kaldor, 1956). Recently, a number ofauthors used agent-based models to explore the link between in-equality and structural change (Ciarli et al., 2012) and between in-equality and financial fragility (Riccetti et al., 2016; Cardaci andSaraceno, 2015) in agent-based models. The main result from thisline of research is that more inequality contributes to macroeconom-ic volatility, and may cause large unemployment crisis and loweroutput growth.

To test how income inequality affects financial and economic stabil-ity in ourmodel, we run 100model simulationswith randomly generat-ed numbers of capital and energy owners in the population. The incomeof capitalists and energy owners is high compared toworkers' wages. Asa result, the larger the frequencies of capital and energy owners in thepopulation are, themore evenly distributed the income is. Interestingly,we could identify five different parameter settings (values of frequen-cies of capital and energy owners out of 100 simulations) which result-ed in no banks' collapses over 1000 time steps. This would suggest thataccumulation of savings by affluent socio-economic groups, underlyingwealth inequality, may be an important source of financial instability.This effect is likely to arise in our model, as accumulation of wealth bycapital owners implies that less income goes to workers in the form ofwages (under the assumption of a constant monetary base). This resultis in line with Riccetti et al. (2016), who find that more inequalityreduces aggregate demand, amplifying the recession, i.e. causing a dropin production and employment.

Results from model simulations show that more capital owners inthe population improves sales, employment and increases the numberof banks surviving in the last period (see Table B4 in the Appendix).The effect of changes in the number of energy owners in the populationon employment and sales are insignificant. The latter result can be ex-plained by the fact that the income of energy owners fluctuates withthe prices of energy and demand for electricity, sometimes fallingbelow and sometimes exceeding wages, whereas the price of capital isalways more expensive than wages in our model. Our results are inline with Piketty (2013), who has argued that the rate of return oncapital in rich countries is generally larger than the rate of economicgrowth, which then will cause wealth inequality to increase over time.

5.2. Energy subsidies

We consider a stylized electricity market with three types of fuelscompeting for adoption by new power plants. The price of fuel 1 isdecreasing over time, the price of fuel 2 is increasing over time (ac-cording to the Brownian motion), while the price of fuel 3 is steadyover time. Fuels differ with respect to operating costs, intensities ofinput use in production, and the cost of installation (see Appendix).In particular, parameters of fuel 1 are calibrated based on coal, fuel2 of gas and fuel 3 based on nuclear energy in the UK. All parameter,with the exception to the cost of installation, are based on empiricalestimates of coal, gas and nuclear energy in electricity production inthe UK as estimated in Safarzynska (2012). The cost of installation ofnew power plants determines which fuels diffuse in the electricitymarket over time. We run 100 simulations for 3 different policies:(1) in the baseline, fuel 3 was the cheapest to install; (2) under Policy2, fuel 2 was the cheapest to install; and (3) under Policy 3, fuel 1 wasthe cheapest to install (means statistics are summarized in Table B4in the Appendix). These three policy scenarios may be indicative ofdifferent potential pathways of change in the electricity sector inthe future. The reader may think of different scenarios as govern-ment interventions, where the government subsidizes the cost of in-stallation of power plants embodying specific fuels. Appendix B.1presents figures, which show electricity prices and the number ofbanks changing over time in illustrative model simulations underdifferent policy scenarios.


The results from model simulations confirm the expectation thatelectricity use is larger the lower the electricity price is (Table B4).In turn, the higher electricity prices translate into fewer banks sur-viving by the end of the simulations. This relates to the positiveand statistical significant relationship between electricity pricesand money in circulations, i.e. total reserves plus total loans inpanel regressions (we do not report these results here). This raisesa question about causation between money creation and electricityprices. On the one hand, high electricity prices may increasemoney in circulations as banks are asked to issue larger loans tocover the cost of more expensive inputs of production. This is inline with Hamilton (2013), who argues that most economic down-turns followed major postwar oil shocks. On the other hand, it hasbeen suggested that money creation by commercial loans could ren-der an increase in the price of energy, destabilizing the economy(Douthwaite, 2012). Our simulation results do not support this hy-pothesis. Using data generated by our model we find that electricityprices are a Granger cause of the total debt, but not vice versa (seeAppendix B.5 for discussion).

5.3. Lending regulations as a response to warning signs of imminent crises

Predicting economic crises has turned out to be extremely difficult.Nevertheless, recent research indicates that there are generic early-warning signals of a complex system approaching a critical threshold(Scheffer et al., 2009). Several papers have been published on earlywarning signals that could give policymakers clues about an upcomingfinancial crisis.Most of the available contributions are based on dynamicstochastic general equilibrium models, which are unable to capture themagnitude offluctuations offinancial indicators, because of their under-lying assumption that prices clear in the equilibrium (Diks et al., 2015).On the other hand, several authors have developed methods to extractsignals of critical slowing down that precedes a transition from time se-ries. One of such indicators is increased variance in the pattern offluctuations.

We find in our model simulations that increasing variance of totaldebt in the economy goes along with a decreasing number of banks inour model simulations. To prevent the collapse of banks, we examinethe impact of the government imposing a policy of adjusting bank lend-ing in response to signs of a financial collapse approaching. In particular,if the growth in variance of total debt increases above 1, we reduce theprobability of new firms entering the market to 0 for 30 consecutivetime steps.Wemeasure the amplitude of changes in variance, by calcu-

lating standard deviation as 1N−1∑

N

t¼1ðyt−μÞ2, where N captures themov-

ing 10 period window, yt is the total debt at time t, and μ is its 10 periodmean.

We run 100 simulations in the presence of weak network effectswith and without the aforementioned policy of adjusting bank lend-ing. We examine the impact of lending regulations in the presence ofweak network effect instead of the baseline scenario, as the weaknetwork effect has proved to be more prone to bank failures. In par-ticular, we find that under the weak network effect, 21(±4) bankslasted until the end of simulations. In the presence of bank's lendingregulations, the number of surviving banks increased to 29(±5). Inaddition, the mean employment increased from 478(±11) to530(±14), and electricity use from 347(±8) to 385(±22). This sug-gests that a policy adjusting bank lending regulations as a responseto signs of banks' failures may contribute to improving both financialand economic stability.

6. Conclusions

In this paper we have proposed a macro-evolutionary model tocapture feedback mechanisms between heterogeneous populations



nξςσ

ττααασ~xpprp

pδηiKψγℒα

αrf

rl

yιDν̂σν̂σ

nnnrc

α

ααα

Table A3Parameter values associated with banks.

nb The initial number of banks 50rIB The interest rate on interbank loans 0.0008


of consumers, producers, power plants and bankers so as to betterunderstand feedback mechanisms between them. To our knowl-edge, we offer the first framework that combines financial withlabor and energy market as well as innovation by heterogeneousfirms. In our model, firms offer products differentiated with respectto quality. They constantly engage in innovation activities: depend-ing on relative input prices, they invest in improvements in electric-ity or labor productivities. Both entrant firms and incumbent firmsthat cannot afford capital expansion ask banks for loans. If a bankhas insufficient liquidity, it attempts to borrow money in the inter-bank lending market. The network of banks evolves as a result ofeconomic activities of firms. On the demand side, we distinguishthree consumer classes: workers, energy and capital owners. Theyengage in product comparisons before purchasing products, and at-tempt to buy goods that are the most popular in their social net-works. Their evolving preferences determine the direction ofinnovative activities of firms.

The novelty of our model lies in combing electricity and financialmarkets. In particular, electricity is an important input in productionof consumer goods in the manufacturing sector. It is produced bypower plants, which embody different energy technologies. Newpower plants enter the market if existing power plants have toclose their production capacities. The size of installed capacity ofnew power plants is determined by the discounted value of suchinvestments.

We find that the stronger the network effect on the consumerside is, the lower the probability of banks' bankruptcies becomes.This can be explained by the fact that where brand loyalty is impor-tant, the debt-to-equity and market shares are good indicators offirms' future profitability. On the other hand, if the network effectis weak, banks have difficulties in assessing riskiness of firms' invest-ments. More firms will then fail to re-pay their loans. This in turn in-creases the risk of cascades of failures under the weak networkeffects compared to the strong network effect. In addition, we findthat if firms invest only in electricity improvements, and not inboth, i.e. electricity and labor improvements, this may lead to energyrebound, so less desirable environmental performance. The reason isthat in the absence of improvements in labor productivity, moreworkers are employed, who demand more products, whose produc-tion is electricity-intensive, which offsets energy savings from im-provements in electricity efficiency.

We further employed our model to study the impact of so-called“sustainability policies” on the economy, which include: the distrib-utive policy; energy subsidies and banks' lending regulations. Wefind that income inequality improves financial stability as well em-ployment in the economy, thus generates a win-win outcome. Onthe other hand, higher electricity prices can push the economyinto a crisis. The more expensive electricity is, the more loans are is-sued in the economy, which increases the probability of cascades ofbank failures. Finally, our results suggest that a policy of adjustingbank lending regulations as a response to signs of an imminent fi-nancial collapse may contribute to circumventing a crisis. This poli-cy responds to the observation a sudden surge in the variance oftotal debt, which precedes a financial collapse in most modelsimulations.

Using a macro-evolutionary model describing interactions with-in and between interrelated networks of technology, finance andenergy systems has proved useful in assessing potential impacts ofmacroeconomic policies and generating novel insights for policy de-bates. This study can be regarded to open up a new direction for re-search that integrates analysis of financial and economic stabilitywith energetic-environmental performance of economies. The rele-vance of such an approach can hardly be overestimated, in view ofthe connected and urgent challenges we face, in particular to simul-taneously reduce unemployment, inequity and greenhouse gasemissions.


Acknowledgments

We would like to thank Gerald Silverberg for his helpful commentsand suggestions. The research was supported by the National ScienceCentre, Poland, grant 2013/08/S/HS4/00254.

Appendix A. Values of model parameters

Tables A1 to A3 below describe parameter values used in the base-line scenario.

Table A1Parameter values associated with manufacturing firms.

edtp

Parameter

crisis-ener://dx.doi.or

Description

gy policy: Macro-evolutionary modelling ofg/10.1016/j.techfore.2016.07.033

Baseline value(min value,max value)

f
The number of firms in time 0 10 Weight attached to demand 0.5 A positive fraction of the maximum quality ~x ;max 0.02 The elasticity of substitution between electricityand capital
0.5

L
Labor productivity 1 (1,2) E Energy efficiency 1 (1,2) K Capital share in production 0.4 L Labor share in production 0.4 E Energy share in production 0.2 qual Improvements in maximum product quality 0.1
;maxt¼0
Maximum product quality in time 0 2
Et = 0
Energy price at time 0 0.3 L Wage 2.5
Capital price
7.5 innov The probability that a firm will undertake
innovation activity
0.05
robentry
The probability that a new firm enters the market 0.06 Depreciation rate of capital 0.02
jt = 0
Markup at time 0 0.01 jt = 0 Initial capital 1000
Parameter in the mark-up function
0.4 Scalar in the cost function 0.25 Parameter in the cost function 0.5
1
The speed of autonomous improvements towardsthe maximum attainable quality
0.002

2
The competence elasticity 0.002 d The interest rate on firms' deposits 0.00005 f The interest rate on firms' loan 0.001
The amount of goods produced by a new entrant
30 Time before a firm is obliged to repay the debt 15
jt = 0F
Initial deposits 0
lt¼0
Labor productivity frontier 0.1
vl
Annual changes in labor productivity frontier 0.0001
et¼0
Energy productivity frontier 0.1
ve
Annual changes in energy productivity frontier 0.0001 suc The probability that a loan is granted 0.9 p
Table A2Parameter values associated with consumers.

w
The number of workers
techn

650
E The number of energy owners 75 K The number of capital owner 25
The interest rate on consumer deposits
0.00001 The parameter describing the maximum elasticitywith respect to product quality among workers
0.375

Among rich consumers
0.5 3 The network elasticity 0.5 4 The snob effect 0.6 jt = 0C Initial deposits 50 D
ology,


T

β The fraction of deposits 0.05R The minimum reserves 50Rjt = 0 Initial reserves 50presell The probability that a bank will sell capital to a new entrant firm 0.9rle The interest rate on loans in the electricity sector 0.001red The interest rate on deposits in the electricity sector 0.00005D/E* The critical value of debt to equity 25

Table A3 (continued)


Table A4Parameter values: electricity market.

Energytechnology j

αααviχσ

pTjpλDηre

Fi

Ti

C

NNR

St

W

Please cite thfinance and e

Description

is article as: Safarzyńska, K., vannergy interactions, Technol. Fore

Fuel1

den Bergcast. Soc

Fuel2

h, J.C.J.M. Change

Fuel3

KjE
Elasticities of substitution 0.452 0.876 0.2 LjE 0.077 0.035 0.07 FjE 0.471 0.089 0.73 t = 0 Initial thermal efficiency 18.25% 18.5% 22.6% − 0.5σj
2
Mean value of changes in fuel prices 0.02 0.06 –
Th
j A standard deviation of changes infuel prices
0.09
0.17 –
Th
fjt = 0 Initial price of fuela 0.58 0.77 0.5
Maximum lifespan
45 40 30
C
oj Operating cost 1.95 0.285 1.37 j Capacity factor 0.8 0.85 0.75 Eit = 0 Initial deposit 8000
p
the markup on electricity price 0.5 The discount rate 0.1 Parameter in the bt function 0.2 g
a We impose a boundary value 1 on fuel prices to prevent unrealistic escalation of pricesover the next 100 year in view of 1990–2007 trends.

Appendix B. Results from model simulations

Table B1Regression results from panel analysiswith differentmacroeconomic indicators as depen-dent variable.

B
Dependent variable: Thefirst difference of:
Total debt
The log ofemployment
The log of(loan loss/loans)

P

P

rst difference of the log oftotal capital

0.10⁎⁎⁎
0.28⁎⁎⁎ −0.01⁎⁎⁎
(0.00)
(0.002) (0.00) me −0.00002 0.00001⁎⁎⁎ 0.00001⁎⁎
(0.11)
(0.00) (0.00) onstant 0.001⁎⁎ −0.004⁎⁎⁎ −0.002⁎⁎⁎
(0.05)
(0.00) (0.00) of observations 98,991 99,800 99,800 of groups 100 100⁎ 100 -sq. within between overall 0.04 0.19 0.01
0.32
0.02 0.01 0.04 0.19 0.01
Note: standard deviation in parenthesis.⁎⁎⁎ Indicates variables significant at the 1% level.⁎⁎ Indicates variables significant at the 5% level.⁎ Indicates variables significant at the 10% level.

Table B2Mean statistics under different scenarios of energy and labor productivities.

The strength of thenetwork effect

Productivityscenarios

The mean ofelectricityconsumption

The mean ofworkers

rong networkeffect

Both productivities
355.99 489.06 (13.68) (21.50)
No change in productivities
539.76 739.63 (14.11) (19.25)
Changes only inenergy productivity

425.05
791.54 (14.73) (21.45)
Low productivity
555.18 762.47 (20.90) (34.33)
eak networkeffect

Both productivities
346.94 478.25 (7.95) (11.59)
No change in productivities
585.58 804.94 (15.27) (20.12)
., Integrated(2016), http

able B2 (continued)

The strength of thenetwork effect

crisis-energy poli://dx.doi.org/10.10

Productivityscenarios

cy: Macro-evolutionary m16/j.techfore.2016.07.03


odelling of t3

The mean ofworkers

Changes only in energyproductivity

448.30
857.67 (7.18) (12.08)
Low productivity
593.19 814.32 (14.41) (18.34)
Note: standard deviation in parenthesis.

Table B3Results from the OLS regression with the frequency of capital and energy owners as inde-pendent variables.

Dependentvariable

Nr. of banksin the lastperiod

The meannumber ofworkers


Mean numberof productssold

e frequency ofcapital owners

67.72⁎⁎⁎
66.80⁎⁎ 85.92⁎⁎⁎ 304.67⁎⁎⁎
(8.26)
(29.76) (22.43) (55.32) e frequency ofenergy owners
-23.85⁎⁎⁎
2.46 8.85 99.37⁎
(8.85)
(31.87) (24.03) (59.26) onstant 34.99⁎⁎⁎ 510.75⁎⁎⁎ 364.79⁎⁎⁎ 397.93⁎⁎⁎
(2.21)
(7.97) (6.01) (14.82) R2
n
0.45 0.05 0.13 0.25 100 100 100 100
Note: standard deviation in parenthesis.⁎⁎⁎ Indicates variables significant at the 1% level.⁎⁎ Indicates variables significant at the 5% level.⁎ Indicates variables significant at the 10% level.

Table B4Mean statistics from 100 simulations for different energy policies.

Energypolicies

Meanlabor

Meanelectricity

Number of banks inthe last period

aseline
489.06 355.98 39 (21.61) (13.75) (4)
olicy 2
495.56 344.97 33 (27.53) (15.42) (5)
olicy 3
504.92 317.40 30 (29.83) (15.05) (6)
Note: standard deviations in brackets.

B.5. Energy policies

Fig. B.1(b) and (c) shows electricity prices and the number of bankschanging over time in illustrativemodel simulations under different en-ergy scenarios. In particular, Fig. B.1(b) illustrates that the electricityprice is the largest under Scenario 3 (where plants embodying fuel 1 dif-fuse over time). Interestingly, the price of electricity here is larger thanin case power plants embodying fuel 2 diffuse in electricity market,resulting in fuel cost to be much lower than the cost of fuel 1(Fig. B.1(a)). This seemparadoxical, yet it can be explained by thehighercapital operating cost of fuel 1 compared to fuel 2 (note that parametersof fuel 1 reflect coal, while those of fuel 2 gas).

To examine causality between electricity prices and debt, weregressed electricity prices on 10 lagged values of electricity pricesand 10 lagged values of the logarithm of total debt in the economy(using 300 observations from our model simulations from differentenergy scenarios), and tested if lagged value of the total debt are asignificant predictor of electricity prices. In addition, we regressedthe total debt on its own lagged values and lagged values of electric-ity prices. We find that we cannot reject the hypothesis that pastvalues of electricity prices predict the total debt in the economy,but neither the reverse causality (at 10% confidence level). This sug-gests that electricity prices are a Granger cause of the total debt, butnot vice versa.

echnology,


(a) Dynamics of fuel prices (b) Electricity price

(c) Number of surviving banks

Fig. B.1. Three energy policies.


References

Acemoglu, D., Ozdaglar, A., Tahbaz-Salehi, A., 2013. Systemic risk and stability in financialnetworks. NBER Working Paper No. 18727.

Akerlof, G.A., 2002. Behavioral macroeconomics and macroeconomic behavior. Am. Econ.Rev. 92, 411–433.

Alesina, A., Rodrik, D., 1994. Distributive politics and economic growth. Q. J. Econ. 109,465–490.

Allen, F., Gale, D., 2000. Financial contagion. J. Polit. Econ. 108, 1–33.Antal, M., van den Bergh, J., 2013. Macroeconomics, economic-financial crisis, and sustain-

ability transitions. Environ. Innov. Soc. Transitions 6, 47–66.Antal, M., van den Bergh, J.C.J.M., 2014. Re-spending rebound: a macro-level assessment

for OECD countries and emerging economies. Energ Policy 68, 585–590.Assenza, T., Delli Gatti, D., Grazzini, J., 2015. Emergent dynamics of a macroeconomic

agent based model with capital and credit. J. Econ. Dyn. Control. 50, 5–28.Battiston, S., Delli Gatti, D., Gallegati, M., Greenwald, B., Stiglitz, J.E., 2007. Credit chains

and bankruptcy propagation in production networks. J. Econ. Dyn. Control. 31,2061–2084.

Battiston, S., Delli Gatti, D., Gallegati, M., Greenwald, B., Stiglitz, J., 2012. Liaisonsdangereuses: increasing connectivity, risk sharing, and systemic risk. J. Econ. Dyn.Control. 36, 1121–1141.

Bikker, J.A., Metzemakers, P.A.J., 2005. Bank provisioning behavior and procyclicality. J. Int.Financ. Mark. Inst. Money 15, 141–157.

Brandt, M.W., Kinlay, J., 2008. Estimating historical volatility. www.investment-analytics.com.

Campbell, J.Y., Cochrane, J.H., 1995. By force of habit: a consumption-based explanation ofaggregate stock market behavior. J. Polit. Econ. 107, 205–251.

Cardaci, A., Saraceno, F., 2015. Inequality, financialisation and economic crises: an agent-based macro model. Working paper 2015–27, OFCE.

Ciarli, T., Lorentz, A., Savona, M., Valente, M., 2012. The role of technology, organisation,and demand in growth and income distribution. Working Paper Series 2012/06. Lab-oratory of Economics andManagement (LEM), Scuola Superiore Sant'Anna, Pisa, Italy.

Cincotti, S., Raberto, M., Teglio, A., 2010. Credit money and macroeconomic instability inthe agent-based model and simulator Eurace. Economics: The Open-Access, Open-Assessment E-J. 4 (26).

Delli Gatti, D., Gallegati, M., Greenwald, B., Russo, A., Stiglitz, J.E., 2009. Business fluctua-tions and bankruptcy avalanches in an evolving network economy. J. Econ. Interac.Coord. 4 (2), 195–212.

Delli Gatti, D., Desiderio, S., Gaffeo, E., Cirillo, P., Gallegati, M., 2011. Macroeconomics fromthe Bottom-up. Springer, Berlin.

Diks, C., Hommes, C., Wand, J., 2015. Critical slowing down as early warning signals for fi-nancial crises? CeNDEF Working Paper 15_04. University of Amsterdam

Dosi, G., Fagiolo, G., Roventini, A., 2010. Schumpeter meeting Keynes: a policy-friendlymodel of endogenous growth and business cycles. J. Econ. Dyn. Control. 34,1748–1767 Elsevier.

Dosi, G., Fagiolo, G., Napoletano, M., Roventini, A., 2013. Income distribution, credit andfiscal policies in an agent-based Keynesian model. J. Econ. Dyn. Control. 37,1598–1625.

Dosi, G., Fagiolo, G., Napoletano, M., Roventini, A., Treibich, T., 2015. Fiscal and monetarypolicies in complex evolving economies. J. Econ. Dyn. Control. 52, 166–189.


Douthwaite, R., 2012. Degrowth and the supply of money in an energy-scarce world. Ecol.Econ. 84, 187–193.

Duesenberry, S., 1949. Income, Saving and the Theory of Consumer Behaviour. CambridgeUniversity Press, Cambridge.

Dynan, K.E., 2000. Habit formation in consumer preferences: evidence from panel data.Am. Econ. Rev. 90, 391–406.

Fagiolo, G., Roventini, A., 2012. Macroeconomic policy in DSGE and agent-based models.Revue de l'OFCE. 124, pp. 67–116.

Fagiolo, G., Roventini, A., 2016. Macroeconomic policy in DSGE and agent-basedmodels redux: new developments and challenges ahead. LEM Papers Series2016/17.

Farmer, D., Foley, D., 2009. The economy needs agent-based modeling. Nature 460,685–686.

Farmer, D., Lo, A., 1999. Frontiers of finance: evolution and efficient markets. Working Pa-pers. Santa Fe Institute (99-06-039).

Frank, R.H., 2008. Should public policy respond to positional externalities? J. Public Econ.92, 1777–1786.

Gabbi, G., Porter, J., Iori, G., Jafarey, S., 2015. Financial regulations and bank credit to thereal economy. J. Econ. Dyn. Control. 50, 117–143.

Gaffeo, E., Gatti, D.D., Desiderio, S., Gallegati, M., 2008. Adaptive microfoundations foremergent macroeconomics. East. Econ. J. 34, 441–463.

Gai, P., 2013. Systemic Risk. The Dynamics of Modern Financial Systems. Oxford Universi-ty Press.

Hamilton, J.D., 2013. Historical oil shocks. In: Parker, R.E., Whaples, R. (Eds.), RoutledgeHandbook of Major Events in Economic History. Routledge Taylor and FrancisGroup, New York, pp. 239–265.

Henriet, F., Hallegatte, S., Tabourier, L., 2012. Firm-network characteristics and economicrobustness to natural disasters. J. Econ. Dyn. Control. 36, 150–167.

Iori, G., Jafarey, S., Padilla, F., 2006. Systemic risk on the interbank market. J. Econ. Behav.Organ. 61, 525–542.

Jackson, T., Victor, P., 2011. Productivity and work in the ‘green economy’. Environ. Innov.Soc. Trans. 1, 101–108.

Janssen, M.A., Jager, W., 2002. Simulating diffusion of green products. Co-evolution offirms and consumers. J. Evol. Econ. 12, 283–306.

Kaldor, N., 1956. Alternative theories of distribution. Rev. Econ. Stud. 23, 94–100.Katz, M., Shapiro, C., 1985. Network externalities, competition and compatibility. Am.

Econ. Rev. 75, 424–440.Laibson, D., 1998. Life-cycle consumption and hyperbolic discount functions. Eur. Econ.

Rev. 42, 861–871.Leibenstein, H., 1950. Bandwagon, snob, and Veblen effects in the theory of consumer de-

mand. Q. J. Econ. 64, 183–207.Lettau, M., Uhlig, H., 2000. Can habit formation be reconciled with business cycles facts?

Rev. Econ. Dyn. 3, 79–99.Ljungqvist, L., Uhlig, H., 2000. Tax policy and aggregate demand management under

catching up with the Joneses. Am. Econ. Rev. 90, 356–366.Lorenz, J., Battiston, S., Schweitzer, F., 2009. Systemic risk in a unifying framework for cas-

cading processes on networks. Working Paper CCSS-09-011.Malerba, F., Nelson, R., Orsenigo, L., Winter, S., 1999. History friendly models of

industry evolution: the computer industry. Industrial and Corporate Change. 8,pp. 3–41.


http://refhub.elsevier.com/S0040-1625(16)30190-1/rf0005





















http://www.investment-analytics.com

http://www.investment-analytics.com














































































Malerba, F., Nelson, R., Orsenigo, L.,Winter, S., 2001. Competition and industrial policies ina ‘history friendly’ model of the evolution of the computer industry. Int. J. Ind. Organ.19, 635–664.

Malerba, F., Nelson, R., Orsenigo, L., Winter, S., 2009. Public policies and changing bound-aries of firms in a “history friendly model” of the coevolution of the computer andsemiconductor industries. J. Econ. Behav. Organ. 67, 355–380.

Nelson, R.,Winter, S., 1982. An Evolutionary Theory of Economic Change. Harvard Univer-sity Press, Cambridge MA.

Neveu, A., 2013. Fiscal policy and business cycle characteristics in a heterogeneous agentmacro model. J. Econ. Behav. Organ. 92, 224–240.

Oltra, O., Saint-Jean, M., 2005. The dynamics of environmental innovations: three stylisedtrajectories of clean technology. Econ. Innov. New Technol. 14, 189–212.

Persson, T., Tabellini, G., 1991. Is inequality harmful for growth? NBERWorking Paper No.3599

Piketty, T., 2013. Capital in the Twenty-First Century. Harvard university Press.Riccetti, L., Russo, A., Gallegati, M., 2016. Increasing inequality, consumer credit and finan-

cial fragility in an agent based macroeconomic model. J. Evol. Econ. 26, 25–47.Rifkin, J., 2011. The Third Industrial Revolution. Palgrave Mcmillan.Safarzynska, K., 2012. Modelling the rebound effect in two manufacturing industries.

Technol. Forecast. Soc. Chang. 79, 1135–1154.Safarzynska, K., van den Bergh, J., 2010a. Evolutionary modelling in economics: a survey

of methods and building blocks. J. Evol. Econ. 20, 329–373.Safarzynska, K., van den Bergh, J.C.J.M., 2010b. Demand-supply coevolution with multiple

increasing returns: policy for system transitions. Technol. Forecast. Soc. Chang. 77,297–317.

Safarzynska, K., van den Bergh, J.C.J.M., 2011. Industry evolutional, rationality and electric-ity transitions. Energ Policy 39, 6440–6452.

Safarzynska, K., Frenken, K., van den Bergh, J.C.J.M., 2012. Evolutionary theorizing andmodelling of sustainability transitions. Res. Policy 41, 1011–1024.

Saint-Jean, M., 2006. Environmental Innovation and Policy: Lessons From an EvolutionaryModel of Industrial Dynamics. (downloadable at www.mnp.nl).

Saunders, F.D., 2008. Fuel conserving (and using) production functions. Energy Econ. 30,2184–2235.

Saviotti, P., Pyka, A., 2008. Product variety, competition and economic growth. J. Evol.Econ. 18, 323–347.

Scheffer, M., Bascompte, J., Brock, W., Brovkin, V., Carpenter, S.R., Dakos, V., Held, H., vannes, E.H., Rietkerk, M., Sugihara, G., 2009. Early-warning signals for critical transitions.Nature 461, 53–59.

Schweitzer, F., Fagiolo, G., Sornette, D., Vega-Redondo, F., Vespignani, A., White, D.R., 2009.Economic networks: the new challenges. Science 325, 422–425.

Sinn, H.-W., 2012. The Green Paradox: A Supply-Side Approach to Global Warming. TheMIT Press, Cambridge, MA.

Sorrell, S., 2007. The rebound effect: An assessment of the evidence for economy-wideenergy savings from improved energy efficiency. UK Energy Research Centrehttp://www.ukerc.ac.uk/Downloads/PDF/07/0710ReboundEffect.


Steinbucks, J., 2010. Interfuel Substitution and Energy Use in the UK Manufacturing Sec-tor. Faculty of Economics and EPRG, University of Cambridge.

Stiglitz, J.E., Gallegati, M., 2011. Heterogeneous interacting agent models for understand-ing monetary economies. East. Econ. J. 37, 6–12.

Stock, J., Watson, M., 1999. Business cycle fluctuations in U.S. macroeconomic time series.In: Taylor, J., Woodford, M. (Eds.), Handbook of Macroeconomics. Elsevier Science,Amsterdam.

Summer, S., Silver, S., 1989. Real wages, employment, and the Phillips Curve. J. Polit. Econ.97, 706–720.

Tedeschi, G., Mazloumian, A., Gallegati, M., Helbing, D., 2012. Bankruptcy cascades in in-terbank markets. PLoS One 7, e52749.

Thurner, S., Polenda, S., 2013. Debrank-transparency: controlling systemic risk in financialnetworks. Nat. Sci. Rep. 3.

Van den Werf, E., 2008. Production functions for climate policy modeling: an empiricalanalysis. Energy Econ. 30, 2964–2979.

van der Ploeg, F., Withagen, C., 2010. Is there really a green paradox? CESifo WorkingPaper Series 2963. CESifo Group Munich

Veblen, T., 1922. The Theory of the Leisure Class: An Economic Study of Institutions. TheMamillan Company, New York.

Windrum, P., Birchenhall, C.T., 1998. Is life cycle theory a special case?: dominant designsand emergence of market niches through co-evolutionary learning. Struct. Chang.Econ. Dyn. 9, 109–134.

Windrum, P., Birchenhall, C.T., 2005. Structural change in the presence of network exter-nalities: a coevolutionary model of technological successions. J. Evol. Econ. 15,123–148.

Windrum, P., Carli, T., Birchenhal, C., 2009a. Environmental impact, quality, and price:consumer trade-offs and the development of environmentally friendly technologies.Technol. Forecast. Soc. Chang. 76, 533–551.

Windrum, P., Carli, T., Birchenhal, C., 2009b. Consumer trade-offs and the development ofenvironmentally friendly technologies. Technol. Forecast. Soc. Chang. 76, 552–566.

World Bank, 2014. Online database (Accessed 15.09.2014).

Karolina Safarzynskaworks as an Assistant Professor at the Faculty of Economic Sciencesat the University of Warsaw. She holds a Ph.D. from VU University Amsterdam. Her re-search concerns evolutionary modelling of transitions to sustainable development.

Jeroen van denBerghworks on the interface of economics, environmental science and in-novation studies. Since 1997 he has been full professor of Environmental Economics at VUUniversity Amsterdam, and since 2007 also ICREA research professor at UniversitatAutònoma de Barcelona. He was a member of the Energy Council of the Netherlandsand is currently editor-in-chief of the journal “Environmental Innovation and SocietalTransitions”. Hewas awarded the Royal/Shell Prize 2002 and IEC's Sant Jordi Environmen-tal Prize 2011.































http://www.mnp.nl










http://www.ukerc.ac.uk/Downloads/PDF/07/0710ReboundEffect

http://www.ukerc.ac.uk/Downloads/PDF/07/0710ReboundEffect

































Date post:	11-May-2021
Category:	Documents
Upload:	others
View:	4 times
Download:	0 times

Technological Forecasting & Social Change · 2017. 10. 16. · Technological Forecasting & Social...

Documents