Business debt dynamics in a micro-founded stock-ow consistent model

L. de Carvalho, C. Di Guilmi
Sao Paulo School of Economics (EESP), Sao Paulo, Brazil. Business School - University of Technology, Sydney. Australia.
Draft version May 20, 2013

Abstract The 2008 nancial turmoil has drawn the attention to the crucial role of credit as a factor leading both to the instability of the system and to a strengthening of real-nancial linkages in the economy. This view was central to the work of Hyman Minsky, who stressed the systemic eects of nancial fragility at the micro-level. This paper introduces heterogeneous microeconomic behavior into a demand-driven stock-ow consistent model, in order to study the joint dynamics of leverage, income distribution and aggregate demand. The distinctive feature is in that the aggregation of heterogeneous agents is not performed numerically as in traditional agent-based models but by means of an innovative analytical methodology, originally developed in statistical mechanics and recently imported into macroeconomics. The new methodology opens an original perspective for economic policy modeling, in particular for addressing situations of coordination failure and excessive leverage.
Corresponding author: Corrado Di Guilmi. University of Technology, Sydney - PO Box 123, Broadway, NSW 2007, Australia. Ph.: +61295147743, Fax: +61295147711. E-mail: corrado.diguilmi@uts.edu.au.

Introduction

The 2008 nancial turmoil itself, and the process of de-leveraging by the private sector observed in the following years, have drawn the attention to the crucial role of credit as a factor leading both to the instability of the system and to a strengthening of real-nancial linkages in the economy. This view, which was central to the work of Hyman Minsky, is also supported by the vast historical evidence presented in Schularick and Taylor (2012), which highlights that credit booms tend to be followed by deeper recessions when compared to other nancial crises episodes. The macroeconomic framework proposed by this paper focuses on the relationships between private sector leverage, income distribution and aggregate demand as a source of instability of the economic system. The theoretical model involves heterogeneous rms aggregated by means of an innovative analytical methodology. As it is well known, economic models rooted in general equilibrium and rational expectations, which have dominated theoretical developments in macroeconomics in the past three decades, have supported the view that markets - albeit being subject to random shocks - are inherently stable, with all uncertainty as exogenous. The nancial crisis has led a larger group of macroeconomists to the understanding of the need to add an (inherently) unstable nancial sector and its relationship with the real side of the economy to their theories. This challenge has boosted the interest toward three main alternative approaches to economic modeling. The rst relates to the incorporation of information asymmetries, sticky prices and bounded rationality into dynamic stochastic general equilibrium (DSGE) models (Smets and Wouters, 2007; De Grauwe, 2010). For representing the economy as a self-regulating system in which market forces lead to a stable equilibrium, the new (DS)GE approaches still do not allow for the incorporation of independent nancial dynamics and the study of systemic risk. Moreover, by still treating macroeconomic relationships as reecting the behavior of its constitutive parts, these models do not allow for the incorporation of fallacies of composition and other emerging properties of complex systems. The stock-ow consistent (SFC) models, rst developed by Tobin (1969) and Godley and Lavoie (2007), have also received renewed attention (Khalil and Kinsella, 2011; Bezemer, 2010, among others). By taking into account all ows of income between dierent sectors in the economy as well as their accumulation into nancial and tangible assets, this type of models were able to trace the 2

ows of credit and stock of debt that could help predict the crisis (Godley, 1999). Besides allowing for formal Minskyan analyses of corporate debt and nancial fragility (Dos Santos, 2005), SFC models have recently been used to study the macroeconomic eects of shareholder value orientation and nancialization (Treeck, 2009), as well as that of household debt accumulation (Kim and Isaac, 2010). The limitation of the SFC approach is that of modeling economic behavior in aggregative terms, thus excluding the heterogeneity of agents as a source of nancial instability. The relevance of a microeconomic analysis in modeling nancial fragility is stressed by Minsky: an ultimate reality in a capitalist economy is the set of interrelated balance sheets among the various units (Minsky, 2008, 116). Indeed one of the distinctive features of his Financial Fragility Hypothesis is the heterogeneity of the nancial conditions of economic units. Taylor and OConnell (1985) remark that shifts of rms among classes as the economy evolves in historical time underlie much of its cyclical behavior. This detail is rich and illuminating but beyond the reach of mere algebra . This lack of proper modeling tools is one of the reasons why Taylor and OConnells model and the vast majority of the literature of Minskyan inspiration, including SFC models, is formulated in aggregative terms. The recent development of new computational and analytical tools has now made feasible the solution of macroeconomic models with heterogeneous agents. Based on the idea that any aggregate economic system is more than the sum of the microeconomic decisions of (rational) agents, the third strategy which gained force after the crisis uses agent-based models to study the emerging properties of decentralized microeconomic interactions taking place in complex and adaptive economic and nancial systems. As argued by Delli Gatti et al. (2010), agent-based models can outperform traditional ones in explaining a wide variety of aggregate phenomena such as uctuating growth, bankruptcy chains and rms sizes and growth rates distributions. In particular, such models can be found to be very useful for the analysis of nancial instability, which clearly requires an understanding of the economy as an out-of-equilibrium system incorporating heterogeneous economic behavior (Delli Gatti et al., 2005, 2010; Ussher, 2008). The agent-based and the SFC approaches to economic modeling can be thought of as complementary in their understanding of the crucial role of real-nancial linkages for the instability of the economic system, as well as its macroeconomic dynamics. Nevertheless, the fact that agent-based models 3

can only be solved numerically has two main drawbacks. The rst one is that, while the micro-behavioral rules are dened and modeled, there is no analytical denition for the relationships between macro and micro-variables. Second, as a consequence the causality links within the system cannot be clearly identied. In this context, this paper introduces heterogeneous microeconomic behavior into a demand-driven SFC model. The model developed here is composed of rms, households and a nancial sector. Firms have heterogeneous degrees of nancial soundness. The distinctive feature is in that the aggregation of heterogeneous agents is performed by means of an innovative analytical methodology originally developed in statistical mechanics and recently imported into macroeconomics (see Aoki and Yoshikawa, 2006; Di Guilmi, 2008; Foley, 1994; Weidlich, 2000, among others). This modeling approach builds from the idea that, as the economy is populated by a very large number of dissimilar agents, an analytical model cannot keep track of the conditions of every single agent at each point in time. As Aoki and Yoshikawa (2006) remark: the point is that precise behavior of each agent is irrelevant. Rather we need to recognize that microeconomic behavior is fundamentally stochastic. Therefore, a microfounded analytical model should look at how many agents are in a certain condition, rather than at which agents, and represent their evolution in probabilistic terms. This approach is particularly suitable to microfound SFC models, since it is able to endogenously derive the macro-equations and the dynamics of ows from the microeconomic behavioral rules, without imposing ad-hoc constraints. In order to implement this method in the stock-ow consistent approach, we rst simulate the model as agent based with full heterogeneity of rms. These simulations serve the only purpose of generating the values of the variables that will be subsequently used for the simulation of the system of ows, as done in standard stock-ow consistent models. The proposed framework has two main objectives. First, the closedform solution will present the analytical links between the nancial microvariables and the macroeconomy, in order to study the joint dynamics of leverage, market capitalization, income distribution and aggregate demand in the cyclical pattern of booms and busts which is typical in nancial fragility literature. Second, the theoretical structure can assess the eects of the interaction between leverage dynamics and income inequality, by studying the shift of 4

households between classes of income along the business and leverage cycle. Whereas in standard macro-models the evolution of the economy is determined by aggregate common shocks, in our framework the idiosyncratic micro-shocks drive the phase transitions of the system. The analysis thus can set the stage for micro and macro policy experiments that will be developed in a further stage. The paper will be structured as follows. Section 2 presents the behavioral equations and the short-run equilibrium of the model. Section 3 species the rms transition dynamics and mean-eld approximation procedure for the analytical solution of the model. In particular in this section we derive the system of equations for the aggregate variables, expressed as functions of rms micro-variables. The dynamics of the micro-, meso- and macrovariables are studied in section 4. Section 5 concludes.

The model

The economy described in this paper is composed by rms, households and a nancial sector. As in the conventional neo-Kaleckian literature, prices are set as a mark-up over labor costs, investment behavior is determined independently and the degree of capacity utilization of rms adjusts to the quantity they sell. The mark-up and the functional distribution of income are assumed to (exogenously) depend on the degree of industrial concentration and the relative bargaining power of workers and capitalists. Firms are divided into two classes and switch between them. While hedge rms nance all their investment with internal resources, borrowing rms nance part or all their investment with stocks and/or bonds. The method proposed in Section 3 will allow the share of rms in each class to aect macroeconomic dynamics. The household sector is also divided into two classes, namely that of wage earners and managers, with the latter receiving a share of rms prots1 . Households allocate their wealth between money and rms shares. Interest rates on bonds are assumed to be set exogenously by the Central Bank, with the price of rms shares being determined by supply and demand in the market for stocks, rather than ows, of these shares. Capital gains (or losses) then aect consumption levels via wealth eects, thus allowing for
There is no microfoundation for the household sector, which is treated as an aggregate with two types of income.
1

the study of the role of asset price booms and bursts in aggregate demand. Finally, the nancial sector is considered as an aggregate: its basic role is to provide loans, hence holding debt (or bonds) as an asset, and to create money deposits endogenously as liabilities.

2.1

The Firms

A single rm is identied by the superscript j , while its state or group by the subscript z = 1, 2. Thus when a variable is written as xj 1 , it refers to the rm j belonging to state 1; a variable with only the subscript indicates the mean-eld value (the average value for the units in the group). Firms variables are represented by small letters, while capital letters indicate macrovariables. The numbers of rms in each group are indicated by N1 and N2 with N1 (t) + N2 (t) = N (t). Firms prefer to nance their investment with internal resources they have previously accumulated in the form of money mj and the ow of retained prots aj . If these are not sucient they issue stocks and bonds. Accordingly, we can dene two classes of rms: z = 1: borrowing rms: that nance part or all their investment with stocks and/or bonds: m j ( t ) + aj ( t ) < i j ( t ) . (1)

z = 2: hedge rms: that nance all their investments with internal resources: m j ( t ) + aj ( t ) i j ( t ) . (2) As for the investment function we use the valuation ratio (Taylor, 2012), which is the ratio between the values of equity and the value of capital assets in the economy. In the present treatment we set it equal to h(t) = P e(t)E (t) K (t) (3)

where P e is the stock price and K the aggregate stock of capital assets. The price of new capital is assumed constant and is therefore normalized to one for simplicity. 6

The investment function for the rm j is given by2

j j ij z (t) = h(t) + z a (t) + u (t)

(4)

where i is investment, u is the capacity utilization ratio and , z , > 0. This formulation recalls the investment functions in Delli Gatti et al. (1993) where the sensitivity to internal nance analytically devices the Minskyan borrowers and lenders risk. In their work the price of equity works as the Tobin q. Following Fazzari et al. (1988) and Delli Gatti et al. (1993), we assume that 1 > 2 , that is borrowing rms are more sensitive to internal nance as they face the risk of bankruptcy and change their behavior in order to minimize it. Equation (4) involves four factors: a macro-eect (h), a meso-eect (z ), a micro-eect (aj ) and a variable combining micro and macro eects (uj ). All rms adopt the same Leontief-type technology with constant coefcients. As a consequence, the demand for labor at full capacity can be residually quantied once the stock of capital is determined by investment decisions in the previous periods. The supply of labor is innitely elastic. Accordingly the production function F gives the potential output q j for rm j q j (t) = F (k j (t), lj (t)) (5) with k and l representing, respectively, physical capital and labor. Even if excess capital may exist, the output-labor ratio, which is indicated by , is constant, so that rms are assumed not to hire excess labor3 . Since technology exhibits xed coecients, it is then possible to dene the potential output only as a function of capital so that q j (t) = 1/ k j (t) where the inverse of the capital productivity is a constant parameter.
For computational needs, in the multi-agent simulations we consider a sequential economy that evolves in discrete time. For this equation investment depends on the previous j j period quantities so that ij z,t = ht1 + z at1 + ut1 . 3 As described in Dutt (1984), the asymmetry which allows excess capital to exist, but the labor-output ratio to be xed technologically can be justied based on the simplifying assumption that labor will not be hired if it does not contribute to production, whereas the stock of capital is determined by previous investment decisions and may be held in excess.
2

(6)

The degree of capacity utilization uj of each rm is dened as the ratio of actual output q j sold by the rm to potential output q j , being equal to one at full capacity and smaller than one with excess capacity, so that uj ( t ) = q j (t) q j (t) = 1 q j (t) k j (t) (7)

Hence, with as a positive parameter, uctuations in the degree of capacity utilization of the rms will track changes in the actual output-to-capital ratio.4 The quantity actually sold by a rm is subject to a preferential attachment shock. It comes from the assumption that the total demand pQ(t) is known, as it amounts to the sum of consumption of capitalists and workers, but its distribution among rms needs to be identied. We assume that this distribution is partially stochastic. In particular, the demand is allocated on the base of the relative of size of rms (proxied by their capital) to which a stochastic idiosyncratic shock s is added. In particular, dening s as a uniformly distributed stochastic variable with E[ s] = 0, we have k j (t) s (t) = s (t) 1 K (t)
j j

(8)

in order for q j to be equal to Q. Accordingly the quantity actually sold by the single rm is q j ( t ) = Q( t ) 1 + sj ( t ) k j (t) K (t) (9)

where Q(t) is the total demand, given by the consumption of managers and wage earners5 . Following Kalecki (1971), rms are assumed to set the price as a mark-up on the cost of labor, while holding excess capacity6 w p = (1 + ) .
4 5

(10)

The capital-output ratio is therefore greater than when there is excess capacity. Formally Q(t) = (C (t) + C (t))/p, where C , C are quantied below. 6 See Rowthorn (1982), Dutt (1984) and Taylor (1985) for early models of growth and distribution in this tradition.

With taken as a parameter, the mark-up rate is constant and the labor share of output is given exogenously 7 = 1 w1 = p 1+ 1+ (11)

The gross prot share of aggregate output will then be given by =1= (12)

Each rms retained prots are computed as the dierence between its gross prots as given by (12), the net interest it pays8 and the portion of net prots (dened as gross prots minus interest payments) which is shared with rms managers. Since the ow of gross prots is given by a constant share of each rms output by the mark-up rule, retained prots of each rm are given by aj (t) = (1 ) pq j (t)[1 + sj (t)] r[bj (t) mj (t)] (13)

A rm fails if aj c, where c is a constant. The case c = 0 corresponds to bankruptcy if the rm is unable to pay interest on bonds without issuing new debt (no Ponzi scheme assumption). The probability for a bankrupted rm of being replaced is directly proportional to the performance of the economy in the previous period. Any excess of retained prots over investment will be held by the rm in the form of money m. The law of motion for the stock of money held by rms is thus given by m j ( t ) = aj ( t ) i j ( t ) (14) Whenever the ow of investment desired by the rm is higher than the stock of money it holds plus its retained prots in the period, the rm will
Again as in Kalecki (1971), the mark-up and the functional distribution of income are assumed to depend on structural characteristics of goods markets such as the degree of industrial concentration and the relative bargaining power of workers and capitalists. Distribution can also be endogenized in the short-run if the power of bargaining of workers is assumed to depend on the unemployment rate or the rate of capacity utilization, possibly giving rise to Goodwin (1967) type of cycles. Such predator-prey dynamics between demand and distribution are studied for instance in Skott (1989) and Barbosa-Filho and Taylor (2006), but will not be allowed for in this version of the model. 8 Firms are assumed to pay interest on bonds b and receive interest on money deposits m at the same interest rate r.
7

seek external nance to cover the dierence. In particular the rms will nance its investment for a share with debt and the rest by issuing new equities. The law of motion for the accumulated stock of rms debt b is then j ( t ) = ij ( t ) aj ( t ) m j ( t ) b 1 1 1 The amount of equities for borrowing rms evolves according to
j j j e j 1 (t) = (1 ) i1 (t) a (t) m1 (t) /P e(t).

(15)

(16)

Borrowing rms that become hedge carry on the quantity of shares previously issued.

2.2

The household sector

The household sector is also divided into two sub-categories, namely that of wage earners (subscript ) and prot earners (subscript ). As described in the previous subsection, workers earn wages w which add up to a constant share of total output Q. Managers (prot earners) receive a share out of rms net prots9 . All households allocate their total wealth W between money M and shares E . Households disposable income Y is composed by wage or prot earnings plus the interest received on the money deposits they hold M , so that Y (t) = pQ(t) + rM (t) Y (t) = [pQ(t) rB (t)] + rM (t) (17) (18)

The wealth of both classes is accumulated as an eect of savings S and capital gain G ( t ) = S ( t ) + G( t ) W (19) where savings S are dened as the dierence between households disposable income and consumption levels S (t) = Y (t) C (t), and G(t ) = [P e(t ) P e(t)]E (t). Finally, consumption spending by each class will be assumed
This distinction is needed to allow for some consumption out of prots. There are two options here. The rst is to have the two types of households with no transition possible between the two. The second is to do the same as for rms, and allow for social mobility. Only the rst option will be developed hereafter.
9

10

to be a xed proportion of both disposable income and the capital gains obtained on equity: C (t) = (1 s )Y (t) (20) C (t) = (1 s )Y (t) + (1 )G(t) (21) where s and s are the propensities to save out of workers and managers disposable income, and is the propensity to save out of managers capital gains. The demand of rms shares money from household is assumed to have the following functional form 1 P e(t)E (t) = W (t) 1 exp[r r G G(t dt)] The demand for money is residually determined as Mh ( t ) = W ( t ) P e ( t ) E ( t ) Given that only prot earners demand for share, we have that ( t ) = Y ( t ) C ( t ) M and, accordingly M ( t ) = Mh ( t ) M ( t ) (25) (24) (23) (22)

2.3

The nancial sector

The nancial sector is considered as an aggregate. It gives loans to rms, hence holding bonds as an asset, and creates money deposits endogenously as liabilities. The interest rate paid on deposits and on loans is considered to be the same for simplicity, so that the nancial sector does not make any prots (net worth is zero). Thus, the total stock of money M held by both households and rms needs to be equal to the stock of bonds B : B (t) =
j

M j ( t ) + M ( t ) + M ( t )

11

2.4

Goods market equilibrium

Total output Q is divided between aggregate consumption C and investment I: pQ(t) = (1 s )Y (t) + (1 s )Y (t) + (1 )G (t) + I (t) After substituting the labor share and the prot share from expressions (11) and (12) in (17) and (18) and solving for Q we obtain 1+ [I (t) + A(t)] s [1 (1 s )]

pQ(t) = where

(26)

A(t) = r[(1 s )M (t) + (1 s )(M (t) B (t))] + (1 )G (t) Table 1 provides a visualization of the balance sheets of the productive, household and nancial sectors while table 2 illustrates the social accounting matrix for our economy10 .

Firms dynamics

As illustrated in sections 1 and 2, the analytical solution method adopted by this project operates through a reduction in the heterogeneity by grouping the agents in clusters. For each cluster a representative agent can be identied. To this aim, in the present treatment we take the average of the relevant micro-variables within the group. This procedure is dened as the mean-eld approximation, which essentially involves reducing the vector of observations of a variable over a population to a single value (Brock and Durlauf, 2001). The following step is presented in subsection 3.1 and concerns the denition of the probabilistic rules for the switching of rms among the dierent groups. The dynamics of the number of agents in each cluster can be represented by a stochastic process of the Markovian type. This class of processes can be analytically described by a master equation which is a stochastic differential equation. The main steps and the outcome of the solution method
Physical capital in rms balance sheets is evaluated at its market price and not at the evaluation price h. We do not therefore consider for the moment the eect of capital gains on rms balance sheets.
10

12

for the master equation are presented is subsection 3.2. In particular, the master equation solution can be expressed in compact form by an ordinary dierential equation plus a stochastic component, given by a Wiener process. Both the ordinary dierential equation and the noise term are formulated as functions of the micro-variables that determine the transition of agents between the dierent groups. This result is then used in subsection 3.3 to derive the laws of motion of the aggregate variables.

3.1

Transition probabilities

The micro-probability for a rm of transitioning from one state to another is determined by its capacity of satisfying conditions (1) or(2). These conditions can be quantied by expressing them as functions of the idiosyncratic shock s, whose distribution is known by assumption. Let us preliminarily introduce the variable z , dening it as z = iz (t) mz (t) [pqz (t) r(bz (t) mz (t)] K (t) . (1 )pqz (t) K ( t ) Kz ( t ) (27)

Using equations (1), (2), (9) and (13), it can be demonstrated that a borrowing rms becomes hedge if s( t ) 1 , and a hedge rm becomes borrowing if s( t ) < 2 . (29) (28)

The rst probability is indicated by while the second by . Accordingly, we can write j ( t ) = P r [ s( t ) 1 ] , (30) j ( t ) = P r [ s( t ) < 2 ] . (31) Assuming s to be uniformly distributed in the interval [0.5, 0.5], we quantify the two probabilities using the cumulative distribution function of s as ( t ) = 1 + 0. 5 ( t ) = 0. 5 2 13 (32) (33)

These probabilities concern the transition of an individual rm from one state to another, thus they can be dened as micro-probabilities. In order to estimate the number of transitions at aggregate level, we need to refer these quantities to the number of rms in each state. The congurational transition rates (Weidlich, 2000) read as + (t) = N1 (t) (t) and (t) = N2 (t) (t) (35) The master equation quanties the dynamics of the the probability of having N1 rms in state 1 in a given instant, assuming that the numbers of rms in the two states evolve according to a jump Markov process. It can be formulated as the balance equation between the aggregate transition to and from state 1 and expressed as dP (N1 , t) = + (t)P (N1 1)(t)+ (t)P (N1 +1)(t)+[+ (t) + (t)] P (N1 )(t) dt (36) (34)

3.2

Master equations solution: stochastic dynamics of trend and uctuations

The method for the master equation introduced by Di Guilmi (2008), developing Landini and Uberti (2008), yields a system of two equations. The rst one is an ordinary dierential equation which describes the time evolution of the trend of the stochastic process. The second one is a partial dierential equation, known as Fokker-Planck equation, whose general solution identies the probability distribution of the uctuations around the drift component11 . This solution techniques split the state variable in two components (as proposed by Aoki, 2002) according to N1 = N m + N s. (37) The factor m is the trend and represents the deterministic component; the variable s is the spread and quanties the stochastic noise around the trend.
11 For a full detail of the solution method we refer the reader to Di Guilmi (2008), Chiarella and Di Guilmi (2011) and Landini and Uberti (2008).

14

The solution derives an equation for each of the components. In particular, the trend evolves according to the following ODE dm = m ( + )m2 , d (38)

where = t/N . The solution for the spread yields the Fokker-Planck equation for the noise, whose stationary distribution is given by the following Gaussian distribution (s) = C exp s2 2 2 : 2 = . ( + )2 (39)

The dynamics of n1 can be therefore described by n1 ( t ) = m ( + )m2 + dV (t) dt (40)

where dV is a stationary Wiener increment and dW is the stochastic uctuation component in the proportion of speculative rms, coming from the distribution (39).

3.3

Mean-eld equations

Having identied the dynamics of the numbers of the two types of rms, we can now use the mean-eld values of each micro variable within the subpopulation of speculative or hedge rms in order to study the dynamics of the aggregate variables. As for the notation, since the population of rms is reduced to just two, the superscript j is no longer needed for the rm-level variables. The total amount of investment is given by (t) = N h(t) + N1 [1 a1 (t) + u1(t)] + N2 [2 a2 (t) + u2 (t)] I (t) = K (41)

Accordingly, the evolution of debt and the amount of equities of the average speculative rm can be quantied by re-dening equations (15) and (16), respectively, in the following way 1 (t) = [i1 (t) a(t) m1 (t)] b e 1 (t) = (1 ) [i1 (t) a(t) m1 (t)] /P e(t). 15 (42) (43)

Consequently, the dynamics of the aggregate debt in the economy is given by (t) = N1 { [i1 (t) a(t) m1 (t)]} B (44) The total number of shares evolves according to ( t) = N1 {(1 ) [i1 (t) a(t) m1 (t)] /P e(t)} E (45)

Analysis

As specied in section 1, the model is initially simulated as agent based with full heterogeneity of agents. In other words, the behavioral rules introduced in subsection 2.1 are applied on N = 1, 000 potentially heterogeneous rms. At each time step the agent based model provides the mean-eld values of the micro-variables. In this stage heterogeneity is reduced to two types of rms (one borrowing and one hedge) by taking the average of the relevant variables within each group of rms. Subsequently, the system composed by the equations introduced in section 3.3 is numerically analyzed. In particular we identify a benchmark scenario and study the dierent dynamics generated by shocking the parameters. In the benchmark scenario the values of the parameters are = 10; = 0.1; 1 = 0.01; 2 = 0.05; = 0.5; s = 0.05; s = 0.7; = 0.75; r = 0.0015; = 0.8; = 1; r = 1; G = 0.000001; N = 1, 000. Figure 1 shows the evolution of aggregate demand, debt and investment for a single simulation. The three variables display the same trend of growth, with short and very regular cycles for demand and investment. The growth of debt is relatively smooth compared to the other series.

4.1

Eects of the heterogeneity of the investment rules

Figure 2 illustrates the evolution of the number of borrowing rms, on a total of 1,000 rms. It adjusts relatively soon to its steady state level of about 400 and then uctuates quite widely. Figure 3 provides some insights about the degree of heterogeneity between the two groups of rms. The upper panel reveals that investment is remarkably larger for hedge rms. In particular, it grows exponentially for hedge rms whereas it reaches a steady state level for borrowing ones. This is mainly due to the higher level of accumulated prot for hedge rms (central 16

panel) that more than compensate for the lower elastiticity of investment to internal nance, since 1 > 2 in equation (4) by assumption. The higher rate of growth for rms in this group generates a larger capacity utilization, as shown by the bottom panel of gure 3. This in turn further increases the investment of hedge rms compared to the borrowing, in a self-reinforcing mechanism. In order to further investigate the eects of heterogeneity we shock the parameters 1 and 2 and look at the eect on the aggregate debt to capital ratio. As shown by gure 4, variations in the coecient for borrowing rms does not substantially alter the evolution and the steady state of the ratio. However, higher values of 1 cause higher levels of the ratio during the adjustment, as one could expect. Figure 5 displays a more interesting picture for the coecient of hedge rms. A larger value of this parameter compared to the benchmark case (2 = 0.01) pushes rapidly the steady state of the debt to capital ratio close to 0. It is worth noting that in the case 1 = 2 = 0.05 the heterogeneity of behavioral rules is eliminated, being the investment equation identical for both type of rms. The impact on the system of 2 is conrmed by gure 6, which reveals considerable larger rates of growth for aggregate demand for a higher sensitivity of hedge rms investment to internal nance. Changes in the parameter have no relevant eects on the share of borrowing rms. It is particularly interesting to assess the eect of heterogeneity on the degree of nancialization of the system, dened here as the ratio between market capitalization and aggregate demand. Figure 7 plots the dynamics of the ratio between the total market capitalization, measured by P e(t) E (t), and the aggregate demand for dierent values of 2 . In the benchmark scenario (with 2 = 0.01), this ratio displays an increasing trend, since the asset price ination in the long run is larger than the rate of growth of the economy. A larger 2 reduces the degree of nancialization of the economy, bringing the ratio to a constant level. This steady state level is lower for larger 2 . For 2 = 2 = 0.05 the rate of growth of the economy converges to the rate of growth of equity price, stabilizing the ratio, despite the larger asset price ination compared to the benchmark scenario. Heterogeneity, and in particular the nancial constraint faced by the borrowing rms, proves to be a factor increasing the dependence of the real sector to the nancial sector. The other two parameters of the investment function have a lesser impact on the dynamics of the debt to capital ratio as illustrated by gures 8 and 9 for, respectively, the elasticity to the evaluation ratio and to the capacity 17

utilization . The plots reveal a slightly higher ratio for higher values of both elasticities.

4.2

Dynamics of leverage

In the benchmark scenario, the ratio between aggregate capital and aggregate debt tends asymptotically to 0 since the accumulation of capital is faster than the accumulation of debt. This outcome can be modied by verying two parameters. The rst one is the propensity to save out of capital gains . Figure 10 shows that the ratio more than doubles for a value of 0.25 compared to 0.5 and 0.75 (benchmark case). A higher propensity to save out of capital gains raises the equity price and, consequently, boosts the level of investment for both hedge and borrowing rms in the same amount. This leads to an increase in the level of debt that is larger than the increase in the stock of capital. Another eect of a higher propensity to save is the rise in the level and in the volatility of aggregate demand (gure 11). This is due to the increase to investment caused by the larger share of income invested in stocks. This also magnes the eects of asset price uctuations on aggregate demand. High level of savings out of capital gains are realistic in a context of growing weight of capital gains on total income. This eect is further discussed below. The accumulation of debt is also impacted by the bankruptcy threshold c. Figure 12 shows that lowering the threshold modies the steady state of the aggregate leverage. The gure does not report the simulation with c = 0 where the ratio spikes to 300. The higher level of debt has no relevant eect on the size of aggregate demand, as shown by gure 13, since hedge rms, which have no debt, account a progressively larger share of aggregate demand as time passes. Hence the levels of investment and demand are comparable to benchmark case whereas aggregate debt is noticeably larger.

4.3

Changes in the propensities to save, distribution of income and preference for liquidity

In general, the eect of an increase in the propensity to save for the dierent categories of income is destabilizing for the system. For a propensity to save of wage earners 0.1, all the rms default at the beginning of 18

the simulation because, in the early stages, the biggest fraction of aggregate demand is provided by salaries. As we show below this is reversed as the simulation runs. A larger propensity to save of prot earners has the eect of lowering the aggregate debt to capital ratio, as shown by gure 14, and increasing level and volatility of aggregate demand, as demonstrated by gure 15. This is due to the fact that an increase in savings raises the demand for equities and their price and, through this channel, the level of investment. The bigger weight of internal nance on the investment decision causes the the volatility of prots to aect more signicantly the volatility of demand. The study of the correlations between the prots and salary bill with aggregate demand highlights the fact that this economy is prot-led. The reliance of the system on nancial income makes growth possible even with a bigger size of the mark-up, and thus a larger share of prots on overall income. Simulations show that the leverage ratio during the transition toward the steady state is lower for larger . This result is due not only to the fact that in our system the only debt is business debt, but also to the increased relevance of capital gains in sustaining growth. In fact, gure 16 reveals that a larger share of income for prot drives up the ratio between size of the equity market and aggregate demand. Also the size of the uctuations of the ratio increases for the higher variance of both equity price and aggregate demand, proving that the system becomes remarkably more volatile. The explanation involves the fact that the propensity to save out of prot is larger than the one for salaries and this increases the demand for equities and, as a consequence, the size of the nancial sector with respect to the real sector. As remarked above in commenting gure 10, in our stylized economy the paradox of thrift is avoided by the increasing nancialization: the increase in investment due to higher evaluation ratio and the fact that consumption is nanced for a progressively increasing weight of nancial income and this avoids the shortfall of a lower propensity to consume out of prots and capital gains. This is consistent with the outcome of the analysis of the sensitivity to the parameter . Our future research will investigate in more detail the conditions under which this process can sustain itself and over which threshold it can lead to a nancial collapse and a depression. Finally, in order to evaluate the role of the preference for liquidity we shock the parameter G , which quanties the sensitivity of investors to capital gains. Results are shown in gure 17. The model reproduces the classical 19

Keynesian result of higher demand and higher accumulation for lower level of preference for liquidity. Also this result is evidently impacted by the faster growth of the nancial sector with respect to the real sector. The three panels show that, at the beginning of the simulation, the patterns generated in the benchmark scenario (G = 106 ) and in the alternative scenario (G = 0.01) are similar. As time passes the two dynamics diverge, with a signicantly higher growth in the alternative scenario.

Concluding remarks

This paper aims to introduce microfoundations in stock ow consistent modeling to study how the interaction between nancial and real sector aects growth and business cycle. To this aim the model is formulated in a bottomup fashion identifying heterogeneous micro-behavioral rules for rms and then deriving the macro-equations for the productive sector using the master equation. The research presented here is at a preliminary stage but nevertheless it provides some insights on the transmission of shocks between the nancial and the real side of the economy. The numerical simulations highlight the relevance of the heterogeneity in the investment function. In particular, if nancially sounder rms are supposed to be less aected by internal nance in their investment decision, the aggregate leverage in the economy is higher in the adjustment phase and can stabilize at a higher steady state. A higher sensitivity to internal nance, even for sounder rms, increases the level of growth of aggregate demand, reducing the dependence of the real sector to the nancial sector. In the benchmark setting, rms are not allowed to roll-over interest on debt. Allowing them to do so makes the aggregate leverage ratio to spike. A larger propensity to save in each of the three income categories (salaries, prots and capital gains) has the eect of noticeably increasing the instability of the system and the size of uctuations. This eect is particularly relevant for the propensity to save out of capital gains, which also appears to be positively related to the aggregate leverage ratio. This result is strictly linked to the growing weight of the nancial sector which aects growth, amplitude of uctuations, distribution of income and role of the preference for liquidity. The next development of this work concerns a more rened study of the conditions under which bubbles and busts are generated in the present 20

setting. Also, the model will be extended by introducing a variable markup, to study the evolution of the shares of income, and the possibility for households to shift between the two categories of prot-earners and incomeearners. From a techinical point of view, future research will involve a deeper analysis of the analytical solution of the model, to be achieved by the study of the dynamical system composed by the dynamic equations for the aggregate variables. The stability properties of this dynamical system will be investigated in order to provide peculiar insights about the role of heterogeneity in the dynamics of ows. The steady state analysis will identify a map of the sustainable and unsustainable long run paths of evolution of the economy as a result of the distribution of leverage and income. It will be possible to appreciate the micro-determinants of the convergent, oscillatory or out of equilibrium dynamics of the economic system. The bifurcation analysis will illustrate the sensitivity of the dynamics to the behavioral parameters of agents, as the decision of rms for investment and of household for consumption.

References
Aoki, M. (2002): Modeling aggregate behaviour and uctuations in economics., Cambridge University Press. Aoki, M. and Yoshikawa, H. (2006): Reconstructing Macroeconomics, Cambridge University Press. Barbosa-Filho, N. and Taylor, L. (2006): Distributive and Demand Cycles in the US-Economy: A Structuralist Goodwin Model, Metroeconomica, 57(3): pp. 389411. Bezemer, D. J. (2010): Understanding Financial Crisis Through Accounting Models, Accounting, Organizations and Society, 35(7): pp. 676 688. Brock, W. A. and Durlauf, S. N. (2001): Discrete Choice with Social Interactions, Review of Economic Studies, 68(2): pp. 23560. Chiarella, C. and Di Guilmi, C. (2011): The Financial Instability Hypothesis: A Stochastic Microfoundation Framework, Journal of Economic Dynamics and Control, 35(8): pp. 1151 1171. 21

De Grauwe, P. (2010): The Scientic Foundation of Dynamic Stochastic General Equilibrium (DSGE) Models, Public Choice, 144: pp. 413443. Delli Gatti, D., Di Guilmi, C., Gaffeo, E., Giulioni, G., Gallegati, M. and Palestrini, A. (2005): A New Approach to Business Fluctuations: Heterogeneous Interacting Agents, Scaling Laws and Financial Fragility, Journal of Economic Behavior and Organization, 56(4): pp. 489512. Delli Gatti, D., Gaffeo, E. and Gallegati, M. (2010): Complex Agent-Based Macroeconomics: a Manifesto for a New Paradigm, Journal of Economic Interaction and Coordination, 5: pp. 111135, 10.1007/s11403-010-0064-8. Delli Gatti, D., Gallegati, M. and Gardini, L. (1993): Investment Condence, Corporate Debt and Income Fluctuations, Journal of Economic Behavior & Organization, 22(2): pp. 161187. Di Guilmi, C. (2008): The generation of business uctuations: nancial fragility and mean-eld interaction, Peter Lang Publishing Group: Frankfurt/M. Dos Santos, C. H. (2005): A Stock-Flow Consistent General Framework for Formal Minskyan Analyses of Closed Economies, Journal of Post Keynesian Economics, 27(4): pp. 712735. Dutt, A. K. (1984): Stagnation, income distribution and monopoly power, Cambridge Journal of Economics, 8(1): pp. 2540. Fazzari, S. M., Hubbard, G. and Petersen, B. C. (1988): Financing Costraints and Corporate Investments, Brookings Papers on Economic Activity, (1): pp. 141195. Foley, D. K. (1994): A Statistical Equilibrium Theory of Markets, Journal of Economic Theory, 62(2): pp. 321345. Godley, W. (1999): Seven Unsustainable Processes: Medium-Term Prospects and Policies for the United States and the World, Economics Strategy Analysis Archive 99-100, Levy Economics Institute.

22

Godley, W. and Lavoie, M. (2007): Monetary Economics: an integrated approach to credit, money, income, production and wealth, Palgrave Macmillan: London. Goodwin, R. (1967): A Growth Cycle, in: Feinstein, C. (ed.), Socialism, Capitalism, and Growth, Cambridge University Press. Kalecki, M. (1971): Selected Essays on the Dynamics of a Capitalist Economy, Cambridge University Press: Cambridge. Khalil, S. and Kinsella, S. (2011): Bad Banks Choking Good Banks: a Simulating Balance Sheet Contagion, Working Papers 201126, Geary Institute, University College Dublin. Kim, Y. and Isaac, A. (2010): The Macrodyamics of Household Debt, Working Papers 1010, Trinity College, Department of Economics. Landini, S. and Uberti, M. (2008): A Statistical Mechanics View of Macro-dynamics in Economics, Computational Economics, 32(1): pp. 121 146. Minsky, H. P. (2008): John Maynard Keynes, McGraw-Hill, rst edition 1975. Rowthorn, R. (1982): Demand, Real Wages and Economic Growth, Studi Economici, 18: pp. 353. Schularick, M. and Taylor, A. M. (2012): Credit Booms Gone Bust: Monetary Policy, Leverage Cycles, and Financial Crises, 1870-2008, American Economic Review, 102(2): pp. 102961. Skott, P. (1989): Eective Demand, Class Struggle and Cyclical Growth, International Economic Review, 30(1): pp. 231247. Smets, F. and Wouters, R. (2007): Shocks and Frictions in US Business Cycles: A Bayesian DSGE Approach, Working Paper 109, National Bank of Belgium. Taylor, L. (1985): A Stagnationist Model of Economic Growth, Cambridge Journal of Economics, 9(4): pp. 383403.

23

Taylor, L. (2012): Growth, Cycles, Asset Prices and Finance, Metroeconomica, 63(1): pp. 4063. Taylor, L. and OConnell, S. A. (1985): A Minsky Crisis, The Quarterly Journal of Economics, 100(5): pp. 87185. Tobin, J. (1969): A General Equilibrium Approach to Monetary Theory, Journal of Money, Credit and Banking, 1(1): pp. 1529. Treeck, T. V. (2009): A Synthetic, Stock-Flow Consistent Macroeconomic Model of Financialisation, Cambridge Journal of Economics, 33: pp. 467 493. Ussher, L. J. (2008): A Speculative Futures Market with Zero-Intelligence, Eastern Economic Journal, 34(4): pp. 518549. Weidlich, W. (2000): Sociodynamics: A Systematic Approach to Mathematical Modelling in the Social Sciences, Gordon and Breach: London.

24

Firm k b m Pe e j

Household sector M + M Pe E W

Financial sector B M

Table 1: Units and sectoral balance sheets

25

Households

Firms

Banks

Total 0 0 0 0 0 0 0 0 0 0 0

Current Capital Consumption C +C Investment +I [N h + N1 (1 a1 + u1 ) + N2 (2 a2 + u2 )] Wages +pQ pQ Prots +(pQ rB ) (N1 a1 + N2 a2 ) (1 )(pQ rB ) + r(N1 m1 + N2 m2 ) Loan interests rB rB Deposit interests r(M + M ) +r(N1 m1 + N2 m2 ) r(M + M + N1 m1 + N2 m2 ) Change in loans +N1 [ (i1 a m1 )] B +M ) Change in deposits (M (N1 m 1 + N2 m 2 M Change in equities P eE +N1 [(1 )(i1 a m1 )] Total 0 0 0 0 Capital Gains P eE +P eE

26

Table 2: Matrix of ows.

Figure 1: Dynamics of aggregate demand, investment and debt.

Figure 2: Dynamics of the number of borrowing rms (total number of rms: 1,000).

27

Figure 3: Investment (upper panel), accumulated prot (central panel) and capacity utilization (bottom panel) for the average borrowing and hedge rms. 28

Figure 4: Dynamics of the debt/capital ratio for dierent values of 1 .

29

Figure 5: Dynamics of the debt/capital ratio for dierent values of 2 .

Figure 6: Dynamics of aggregate demand for dierent values of 2 .

30

Figure 7: Dynamics of equity value to aggregate demand ratio for dierent values of 2 .

Figure 8: Dynamics of the debt/capital ratio for dierent values of .

31

Figure 9: Dynamics of the debt/capital ratio for dierent values of .

Figure 10: Dynamics of the debt/capital ratio for dierent values of .

32

Figure 11: Dynamics of aggregate demand for dierent values of .

Figure 12: Dynamics of the debt/capital ratio for dierent values of c.

33

Figure 13: Dynamics of debt, investment and aggregate demand for c = 0.01.

Figure 14: Dynamics of the debt/capital ratio for dierent values of s .

34

Figure 15: Dynamics of aggregate demand for dierent values of s .

Figure 16: Dynamics of equity value to aggregate demand ratio for dierent values of .

35

Figure 17: Dynamics of debt to capital ratio (upper panel), capital accumulation (central panel) and aggregate demand (bottom panel)for dierent values of G . 36