Kinetic Market Model: An Evolutionary Algorithm

Evandro Luquini,Nizam Omar

DOI: https://doi.org/10.48550/arXiv.1906.01241

2019-06-04

Abstract:This research proposes the econophysics kinetic market model as an evolutionary algorithm's instance. The immediate results from this proposal is a new replacement rule for family competition genetic algorithms. It also represents a starting point to adding evolvable entities to kinetic market models.

Neural and Evolutionary Computing,General Finance

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to understand and apply kinetic market models in economics as an evolutionary algorithm. Specifically, the author proposes a new replacement rule for family competition genetic algorithms, which is based on the mechanisms of kinetic market models. In addition, the paper also explores how to add evolvable entities to kinetic market models to better simulate economic inequality phenomena and explore optimization heuristic methods through these models.

### Main contributions of the paper:
1. **New replacement rule**: The paper proposes a new replacement rule for improving family competition genetic algorithms. This rule allows the model to select and recombine without bias towards the best individuals, while only allowing challenges to the current parental positions, which is different from traditional genetic algorithms.

2. **Kinetic market models as evolutionary algorithms**: The author regards kinetic market models as an instance of an evolutionary algorithm. In this way, the dynamic characteristics of these models can be more deeply understood and applied to optimization problems.

3. **Diversity maintenance**: The experimental results show that the proposed kinetic market replacement rule can better maintain population diversity during the simulation process, while other methods lose diversity earlier.

### Formula explanations:
- **Basic exchange rule of kinetic market models**:
\[
m_i(t + 1)=\epsilon\cdot m_i(t)\quad\text{where}\quad\epsilon = U(0, 1)\quad(1a)
\]
\[
m_j(t + 1)=m_j(t)+(m_i(t)-m_i(t + 1))\quad(1b)
\]

- **Equivalent random search process**:
\[
S = \{s_1, s_2, s_3,\ldots, s_k\}\quad\text{from}\quad U(0, m_i(t))\quad(2a)
\]
\[
m_i(t + 1)=X,\quad\text{where}\quad P(X = x)=\frac{1}{|S|}\quad(2b)
\]
\[
m_j(t + 1)=m_j(t)+(m_i(t)-m_i(t + 1))\quad(2c)
\]

- **Abstract kinetic market model**:
\[
S=\{s_1, s_2,\ldots, s_k\}\quad\text{from}\quad f(m_i(t), m_j(t))\cup\{m_i(t), m_j(t)\}\quad(3a)
\]
\[
W = \{w\in S\mid w\geq m_j(t)\}\quad(3b)
\]
\[
L = \{l\in S\mid l\leq m_i(t)\}\quad(3c)
\]
\[
m_i(t + 1)=X,\quad\text{where}\quad P(X = x)=\frac{1}{|L|}\quad(3d)
\]
\[
Q=\{q\in W\mid q\leq m_j(t)+(m_i(t)-m_i(t + 1))\}\quad(3e)
\]
\[
m_j(t + 1)=\max(Q)\quad(3f)
\]

### Experimental results:
The paper verifies the effectiveness of the proposed kinetic market replacement rule by comparing the performance of different replacement rules (such as elite recombination and best - random selection) on the eil76 problem in TSPlib. The experimental results show that the kinetic market replacement rule in maintaining