2024 Volume 33 Issue 7
Article Contents

Jun Meng, Xia Zhou, Shaoyong Lai. A wealth distribution model with a non-Maxwellian collision kernel[J]. Chinese Physics B, 2024, 33(7): 070501. doi: 10.1088/1674-1056/ad3dc6
Citation: Jun Meng, Xia Zhou, Shaoyong Lai. A wealth distribution model with a non-Maxwellian collision kernel[J]. Chinese Physics B, 2024, 33(7): 070501. doi: 10.1088/1674-1056/ad3dc6

A wealth distribution model with a non-Maxwellian collision kernel

  • Received Date: 21/11/2023
    Accepted Date: 19/03/2024
    Available Online: 01/07/2024
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Figures(2)

Article Metrics

Article views(155) PDF downloads(0) Cited by(0)

Access History

A wealth distribution model with a non-Maxwellian collision kernel

Abstract: A non-Maxwellian collision kernel is employed to study the evolution of wealth distribution in a multi-agent society. The collision kernel divides agents into two different groups under certain conditions. Applying the kinetic theory of rarefied gases, we construct a two-group kinetic model for the evolution of wealth distribution. Under the continuous trading limit, the Fokker–Planck equation is derived and its steady-state solution is obtained. For the non-Maxwellian collision kernel, we find a suitable redistribution operator to match the taxation. Our results illustrate that taxation and redistribution have the property to change the Pareto index.

1.   Introduction
  • The study of wealth distribution has a long history. Since Pareto’s discovery at the end of the 19th century that the wealth distribution in Western countries is characterized by a power law,[1] scholars have applied various techniques to construct evolutionary models of wealth distribution in an attempt to find a universal form of wealth distribution and explain what factors affect the wealth distribution in an economy. About two decades ago, scholars began to study wealth distribution in multi-agent systems with the help of the well-established kinetic theory of rarefied gases, and achieved good results. The key to apply the theory is to construct the Boltzmann equations for the evolution of wealth distribution, which is usually referred as the kinetic equations, which essentially describe the change of agents’ wealth over time.[2]

    The construction of the Boltzmann equation for the evolution of wealth distribution depends on the interaction rules and the frequency of interactions (i.e., collision kernels) between agents. Cordier et al.[3] employed a constant interaction frequency (i.e., the Maxwellian collision kernel, also called the constant collision kernel) and an interaction rule incorporating the savings propensity and the market risk to construct the Boltzmann equation for the evolution of wealth distribution, obtaining that the wealth distribution at the equilibrium state is an inverse gamma distribution. Bisi et al.[4] introduced taxation into the interaction rule, find a suitable redistribution operator, construct a kinetic model containing taxation and redistribution by using a constant collision kernel, and discuss the effects of taxation and redistribution on wealth distribution. Pareschi and Toscani[5] introduced knowledge into the interaction rules, and constructed a kinetic model of wealth distribution with a constant collision kernel to examine the effect of knowledge on wealth distribution. There are many works by using constant collision kernels to construct kinetic models of wealth distributions (see Refs. [6–10]).

    Recently, Furioli et al.[11] pointed out that the constant collision kernel may ignore some behaviors of agents, and the wealth distribution model constructed with the collision kernel is somewhat different from reality. To overcome this deficiency, Furioli et al.[11] proposed a non-Maxwellian collision kernel (variable collision kernel), i.e.,

    where constants κ > 0 and 0 < δ ≤ 1, v and w represent the wealth of two different agents. The collision kernel (1) is used to exclude the possibility that agents with no wealth could enter the market and enhance transactions in which the amount of money of both agents is noticeable. The results in Ref. [11] state that the constant δ changes the Pareto index of the steady-state solution and demonstrate that the solution of the resultant Fokker–Planck equation converges exponentially to equilibrium state. Subsequently, Del-Mul[12] extended the collision kernel (1) to the following form:

    where constants 0 < α ≤ 1,0 < β ≤ 1, and κ > 0. Indices α and β represent certain collective behavior of agents, such as the intensity of willingness to trade. If α = β, equation (2) becomes the collision kernel (1). Moreover, if α = β = 0, the collision kernel (2) becomes a Maxwellian collision kernel. When α = β, Del-Mul[12] obtained the same result as that in Ref. [11]. However, Del-Mul[12] did not discuss the effect of the collision kernel (2) on the wealth distribution when α ≠ β. Ballante et al.[13] examined the wealth distribution of agents with uncertain trading propensities by using the collision kernel (1). Although collision kernel (1) is used in Refs. [11,13] to model wealth distribution, the collision kernel (1) implies that the intensity of agents’ willingness to trade is the same, and thus it does not produce a wealth distribution characterized by bimodal peaks.

    The feature of the collision kernel (2) is that the intensity of each agent’s willingness to trade affects the frequency of interaction. Low transaction intensity assigns a low frequency of interaction, while high transaction intensity assigns a high frequency of interaction. For this reason, different transaction intensities divide agents into different groups. Our work is to consider the wealth distribution from different agent groups and analyze the influence of different transaction intensities on wealth distribution. Our results illustrate that when two groups of agents differ largely in the intensity of their willingness to trade, it leads to a bimodal shape in the steady-state distribution of wealth. This means that if there are two completely different types of agents in a group, some of them prefer a high interaction frequency and others prefer a low one. Then the wealth distribution of the group possesses bimodal. The bimodal distribution of the actual data has been observed in the income distribution of Argentina during the economic crisis.[14]

    From the work in Refs. [4,7], it is known that the wealth kinetic model with taxation and redistribution associated with the Maxwellian collision kernel is able to derive the statistical laws of wealth distribution and analyze the impact of taxation and redistribution on wealth distribution. Inspired by the ideas in Refs. [4,7], we employ a non-Maxwellian collision kernel (2) to construct the wealth distribution model with taxation and redistribution. A redistribution operator associated with the non-Maxwellian collision kernel is analyzed. The results confirm that under the non-Maxwellian collision kernel, taxation and redistribution are able to modify the Pareto index in the steady-state solution and regulate inequality in the distribution of wealth.

    The main contributions of our work are twofold. Firstly, we construct a kinetic model about the evolution of wealth distribution of two groups by using the non-Maxwell collision kernel (2) and obtain a mechanism for the bimodal formation of wealth distribution, i.e., when there is a large difference between the intensities of willingness to trade in the two groups (which leads to a large difference in the frequency of interactions), the steady-state wealth distribution takes on the shape of a bimodal peak. This mechanism is different from those used in Refs. [6,8] for the formation of bimodal distributions. Using the Maxwellian collision kernel, During and Toscani[6] considered the effect of differences in saving propensities between different countries (or groups) on wealth distribution, and Dimarco et al.[8] examined the impact of differences in saving propensities of different groups on wealth distribution when the COVID-19 is prevalent. The results in Refs. [6,8] showed that the wealth distribution is bimodal when differences between saving propensities of different groups are sufficiently large. Secondly, we construct a kinetic model of wealth evolution with taxation and redistribution by using the non-Maxwellian collision kernel (2). For the aim of taxation, we find a redistribution operator corresponding to Eq. (2), which is different from the redistribution operators associated with Maxwellian collision kernels in Refs. [4,7].

    The organizational structure of this paper is as follows. In Section 2, employing the collision kernel (2), we construct a two-group kinetic model for the evolution of wealth. Utilizing the continuous trading limit, we obtain steady-state solutions under certain assumptions. In Section 3, we discuss the influence of taxation and redistribution on wealth distribution under the collision kernel (2). The conclusion is stated in Section 4.

2.   A two-group kinetic model for the evolution of wealth
  • According to the kinetic theory,[2] the evolution of wealth distribution depends on the microscopic interactions between agents involved in the exchange of wealth. Agents participate in transactions with part of their wealth and each transaction is viewed as an interaction. The wealth after the interaction is usually assumed to be nonnegative.

  • As stated in the introduction, agents are divided into two groups according to their different transaction intensities. That is, in the collision kernel (2), α = β means that there is only one type of agent in the market, and α ≠ β indicates that there are two groups of agents in the market. We denote by f1(w,t) and f2(w,t) (w ∈ R+), the wealth distribution at time t ≥ 0 of agents with transaction willingness intensity of α and β, respectively. The total wealth distribution is then given by

    In general, the density function ftot(w,t) satisfies ∫R+ftot (w,t) dw = 1.

    Following the binary trading rule in Ref. [3], when two agents with wealth v and w participate a trade, the post trade wealth becomes

    where λ ∈ (0,1) is the transaction rate parameter. The quantities η1 and η2 are independent and identically distributed random variables with mean zero and finite variance σ > 0. To make the post-interaction wealth non-negative, it is usually assumed that ηi ≥ λ – 1 (i = 1,2). Using binary trading rules (4) and assumption of random variables η1 and η2, we acquire

    implying that the total wealth is conserved in a statistical sense, where symbol 〈⋅〉 indicates the mathematical expectation.

    Utilizing the standard methods of kinetic theory,[2] we obtain that the evolution of wealth distribution fi(w,t) (i = 1,2) satisfies the Boltzmann equation

    where the collision operator

    represents transactions from different groups, and

    represents transactions among agents in the same group. In Eqs. (6) and (7), (v,w) denote the pre-transaction pair that generates the post-transaction pair (v*,w*) according to the interaction rule (4), while the collision kernels Kij(v,w) and Kii(v,w) are given by

    In general, the weak form of the Boltzmann equation is adopted, i.e., for any smooth function ϕ(w) having supported set in R+, the evolution of fi(w,t)(i = 1,2) follows the equation

    For more details on the application of the Boltzmann equation, the reader is referred to Refs. [15–17].

    In Eq. (8), when ϕ(w) = 1, mass conservation is obtained, meaning that the following equation holds:

    where the positive constants ρ1 and ρ2 satisfy ρ1 + ρ2 = 1. If ϕ(w) = w, utilizing Eq. (4), for any Kij(v,w) and Kii(v,w), we obtain

    indicating that the average wealth is conservation. For the sake of the subsequent discussion, we assume that first three moments of the solution fi(w,t) of Eq. (8) are bounded.

  • Generally speaking, it is very difficult to solve the Boltzmann equation directly. For this reason, the Fokker–Planck equation is used as its approximate equation to analyze the large-time behavior of transaction. It is assumed that most transactions correspond to a very small exchange of money, that is, in binary trading rule (4), the transaction rate parameter λ is selected to be small enough, but at the same time it keeps certain traces of the phenomena that affect the market rules at the macro level such as market risk, taxation, and redistribution. This kind of asymptotic analysis is called “continuous trading limit” (see Refs. [3,7,18]). It is achieved by introducing the scaling

    where ξ is a small positive parameter.

    Using expression (9) yields

    For a smooth function ϕ(w), utilizing the Taylor series ϕ(w*) around ϕ(w), we obtain

    where Rξ(v,w) denotes the remainder term including ξn (n ≥ 3/2).

    Inserting Eq. (10) into Eq. (8) gives rise to

    where

    Now, we set

    which implies g1(w,0) = f1(w,0) and g2(w,0) = f2(w,0). Letting ξ → 0, the remainder R∼ξ(v,w)→0 (see Ref. [19]), we obtain

    In Eq. (12), integration by parts yields

    where

    Note that the first subscript of mϵ, j represents the ϵ-th moment of gi(w, τ), and the second subscript denotes the group to which the agent belongs.

    Consequently, the evolution of g1(w, τ) satisfies the Fokker–Planck equation

    In the same way, we have

    In Eq. (13), because the diffusion term and drift term contain mα,1(τ) and m1+α,1(τ), it is difficult to obtain the analytical solution of Eq. (13). In Eq. (14), the same is true. If we assume Q(fi, fi)(w) = 0, that is, agents from the same group do not trade, equations (13) and (14) become

    respectively.

    For 0 < β ≤ 1, according to the Jensen’s inequality

    we have

    If the initial average wealth is not zero, it is known from the conservation of average wealth that the average wealth will not be zero at each subsequent moment. Therefore, we obtain mβ,2(τ) ≠ 0. mα,1(τ) ≠ 0 is derived by the same method.

    The steady-state solution of Eq. (15) is determined by the following equation:

    Using ∫R+g1(w,t)dw = ρ1 and solving Eq. (17), we obtain

    where m∼β=m1+β,2(∞)mβ,2(∞). From Eq. (16), employing the same calculation, the steady-state solution g2,∞(w) is given by

    where m∼α=m1+α,1(∞)mα,1(∞).

    When α = β, equations (15) and (16) are the Fokker–Planck equations corresponding to the bilinear model in Ref. [11]. For more research on the application of the Fokker–Planck equation and its large time behavior, please refer to Refs. [19–21]. In Eq. (18), if w → +∞, we have g1,∞(w) ∼ w−2–α–2λ/σ. The Pareto index is 1 + α + 2λ/σ, implying that wealth inequality for the first group decreases as α increases. Similarly, from Eq. (19) we obtain that wealth inequality for the second group decreases with increasing β. In this case, the total wealth distribution is

    which has a bimodal distribution. Indeed, considering a simple case of m∼β=m∼α=m∼, ρ1 = ρ2 = ρ and letting μ = 2λ/σ, formula (20) becomes

    Letting y=w/(μm∼), we obtain

    The extreme cases in Eq. (22) depend on the solution of the equation

    Clearly, y = 0 is an extremum (a minimum). If y > 0 and β > α, the remaining extreme cases are determined by the equation

    Using the expression of φ(y), we obtain φ(0) > 0 and φ(+∞) < 0. Therefore, the curve z = φ(y) touches the axis z = 0 either in a single point or in three points or others. If there is only one maximum, gtot,∞(w) presents a single peak shape (see Fig. 1(b)). When it has three points, two maximum and one minimum are obtained except y = 0 and gtot,∞(w) shows a bimodal distribution (see Fig. 1(a)). As Ballante et al.[13] stated, the bimodal distribution of wealth occurs in times of poor economic conditions, such as the income distribution due to the Argentinean economic crisis around the year 2002,[14] and the wealth distribution under the influence of COVID-19 epidemic in recent years.[8] From our analysis and the results in Refs. [6,8], it is known that the emergence of the bimodal wealth distribution is the inconsistency in agents’ behaviors in the system due to the shocks to the economy caused by some unforeseen events, resulting in the appearance of different transaction types of agents, such as some people with a high frequency of interactions (or high savings rate) and some people with a low frequency of interactions (or low savings rate). In this case, parts of the agents are extremely rich and some are extremely poor, and the total wealth distribution in the system shows a bimodal peak.

3.   Taxation and redistribution under non-Maxwellian collision kernel
  • Bisi et al.[4] introduced taxation into binary trading rules, selected appropriate redistribution operators under the Maxwellian collision kernel, and obtained that taxation and redistribution affect the Pareto index. Bisi[7] reviewed and studied a kinetic model with taxation and redistribution of wealth. In this section, under the non-Maxwellian collision kernel (2), we introduce taxation into the transaction rule and seek an appropriate redistribution operator to analyze how taxation and redistribution policy affect the Pareto index.

    As it is known, the main function of taxation and redistribution is to adjust the wealth gap and decrease the inequality of wealth distribution. In terms of narrowing the gap between the rich and the poor, the rich should be taxed more and the poor should pay less, but economists are more concerned about which taxation rate can maximize social welfare. Mankiw et al.[22] pointed out that the “flat taxation” (that is, the same taxation rate relies on different income levels) may be close to optimal.

    Following the idea in Ref. [22], we introduce flat taxation into the transaction rule, i.e.,

    where w and w* denote pre- and post-interaction wealth of the agent, respectively. 0 < λ < 1 is a transaction rate parameter, 0 < s < 1 is flat taxation rate, η is a central random variable with mean zero and finite variance σ > 0. To ensure that the wealth after trade is positive and the taxation rate does not exceed the transaction rate, we assume s ≤ min{(1 – λ + η), λ} (see Ref. [4]). In the interaction rule (25), the first term indicates the remaining wealth after the agent participates in market trading and pays taxes. The second term represents the wealth obtained by agents for trading, where v ∈ R+ denotes the wealth owned by agents in the market with a certain distribution H(v). The third term implies the wealth obtained by agents due to risks in the market.

    Denote by f(w,t) the wealth distribution of agents with wealth w > 0 at the moment t ≥ 0. According to the kinetic theory,[2] the evolution of f(w,t) obeys the Boltzmann equation

    where Q(f, f)(w, t) is a collision operator which describes the change of f(w, t) under non-Maxwellian collision kernel (2) and linear transaction rule (25). The weak form of Eq. (26) is generally adopted, that is, for any smooth function ϕ(w), the evolution of f(w, t) follows the equation

    where K(v, w) is the collision kernel defined in Eq. (2). Set Mα=∫R+vαH(v)dv<+∞ (0 ≤ α ≤ 4), mγ(t)=∫R+wγf(w,t)dw (γ > 0) represents the moment of order γ of f(w,t).

    In Eq. (27), we choose ϕ(w) = w and obtain that the average wealth evolution in the presence of taxation satisfies the equation

    Utilizing the Jensen’s inequality, we get

    which results in the inequality

    meaning that the average wealth is bounded if its initial average wealth is bounded. In Eq. (25), If s = 0, i.e., agents do not pay taxes, the evolutionary equation for average wealth is

    According to the idea in Ref. [7], the wealth obtained through taxation is returned to agents by appropriate redistribution policy. In such a way, the total wealth remains unchanged. The amount of wealth owned by agents after trade under collision kernel (2) is different from the amount of wealth owned by agents after trade under constant collision kernel. Furthermore, comparing Eqs. (28) and (29), it is found that the effect of taxation on the evolution of average wealth is captured by the term −κsMαm1+β(t), where m1+β(t)=∫R+w1+βf(w,t)dw. Thus, to ensure matching the collision kernel (2), we use the redistribution operator

    where

    The function Ψ(w) denotes the fraction of the obtained taxation that is distributed to each agent with wealth w, also called redistribution function, and its value could be positive or negative. When it is negative, meaning that a certain part of the population is further taxed. If β = 0, then R(f)(w,t) and Ψ(w) are the redistribution operators and redistribution functions in Ref. [7], respectively. χ in the redistribution function is a free parameter, which represents the redistribution policy and is given by policy maker. The meaning of redistribution policy is consistent with that in Ref. [7] if β = 0. The meaning of redistribution policy should be determined according to specific values of β if β ≠ 0. Because β takes in the interval (0,1], we only analyze the meaning of redistribution policy when β = 1 and Ψ(w) < 0, and the analysis of other β values is similar. When χ < –1, for w∈I1=(0,χ+1χm2(t)), we have Ψ(w) < 0. Therefore, the agents with wealth w ∈ I1 supply additional resources to the agents with w ∉ I1. When χ > 0, for w∈I2=(χ+1χm2(t),+∞), we have Ψ(w) < 0. In this case, the agents with wealth w ∉ I2 are favored, while the agents with w ∈ I2 are taxed more.

    Assume that f(w,t) satisfies boundary condition f(0,t) = 0. The redistribution operator satisfies following constraints:

    and the redistribution function fulfills

    Now we consider the kinetic model with taxation and redistribution mechanism. That is, for the non-Maxwellian collision kernel (2), the linear trading rule (25), and the redistribution operator (30), the evolution of f(w,t) satisfies the Boltzmann equation

    where Q(f, f)(w, t) is a collision operator. For any smooth function ϕ(w), the weak form of the Boltzmann equation with taxation and redistribution becomes

    where the second term on the right hand in Eq. (33) is obtained by integration by parts for R(f)(w,t), bearing in mind that f(0,t) = 0. The redistribution operator has been constructed to compensate the losses of wealth caused by taxation. For this reason, m1+β is a constant (independent of t) in Eq. (33).

    Similarly, following the idea of continuous trading limit, we use the Fokker–Planck equation as an approximation of the Boltzmann equation (33) to discuss the large-time behavior of wealth distribution. For a very small positive parameter ξ ≪ 1, we set λ → ξλ, s → ξs, and η→ξη. Then a second-order Taylor expansion of ϕ(w) around w yields

    Inserting Eq. (34) into Eq. (33), and letting τ = ξt, g(w, τ) = f(w,t), we obtain

    where

    When ξ → 0, R∼sξ(v,w)→0. Then, equation (35) becomes

    Integrating the right side of Eq. (36) by parts and combining with boundary conditions, we acquire the Fokker–Planck equation

    To facilitate analysis, we set σκMα/2 = 1 and M1+α/Mα = m. Obviously, κMα = 2/σ. Thus, equation (37) becomes

    When τ → ∞, we get the steady-state solution g∞(w) satisfying the equation

    Solving Eq. (39), we obtain

    where C satisfies ∫R+g∞(w)dw=1.

    In Eq. (40), if s = 0 or χ = –1, The steady-state solution g∞(w) is the equilibrium wealth distribution without the presence of taxation and redistribution. When w → +∞, we have g∞(w)∼w−(2+β+2λσ+2s(χ+1)σ). Thus, the Pareto index is 1+β+2λσ+2s(χ+1)σ. This result demonstrates that the taxation and redistribution affect the Pareto index under the non-Maxwellian collision kernel (2). Figure 2(a) gives a graph of the steady-state solution g∞(w) as the taxation rate s varies. It is observed that when the rich are taxed (i.e., χ > 0), the inequality of wealth distribution decreases as the taxation rate increases. Figure 2(b) shows a picture of the steady-state solution g∞(w) as the redistribution parameter χ varies. It is seen that, if other parameters in the system do not change, the inequality of wealth distribution decreases with an increase of redistributive parameter χ. According to the above analysis, the choice of redistribution policy would affect the value of redistribution function. Which type of agents (the poor or the rich or others) will benefit should be considered by policy makers.

4.   Conclusion
  • In this work, we employ a non-Maxwellian collision kernel to study the evolution of wealth distribution in a multiagent society. According to the characteristics of the collision kernel (2), we construct a two-group kinetic model for the evolution of wealth distribution. Using the continuous trading limit, we obtain a corresponding Fokker–Planck equation. Under certain assumptions, we derive a steady-state solution, illustrating that the total wealth distribution presents a bimodal shape when different groups of agents have significantly different intensities of willingness to trade. In addition, under the non-Maxwellian collision kernel (2), we find the redistribution operator that matches the taxation and construct a wealth distribution model with taxation and redistribution. The results show that taxation and redistribution have the property to change the Pareto index under the non-Maxwellian collision kernel (2).

Figure (2)  Reference (22)

Catalog

    /

    DownLoad:  Full-Size Img  PowerPoint
    Return
    Return