Abstract:Network economics is the study of a rich class of equilibrium problems that occur in the real world, from traffic management to supply chains and two-sided online marketplaces. In this paper we explore causal inference in network economics, building on the mathematical framework of variational inequalities, which is a generalization of classical optimization. Our framework can be viewed as a synthesis of the well-known variational inequality formalism with the broad principles of causal inference
What problem does this paper attempt to address?
This paper attempts to solve the problem of causal inference in network economics. Specifically, it combines variational inequalities (Variational Inequalities, VI) and causal inference models and proposes a new framework - causal variational inequalities (Causal Variational Inequalities, CVI) to model and analyze causal relationships in network economic systems more accurately.
### Background and Problem Description of the Paper
**Network economics** studies a rich class of equilibrium problems in the real world, which cover multiple fields from traffic management to supply chains and two - sided online markets. Traditional variational inequalities are a generalized optimization method and can be used to model and solve various equilibrium problems, such as convex optimization, non - cooperative games, fixed - point equation solving, and complementary problems. However, traditional variational inequalities have limitations in dealing with causal relationships.
**Causal inference** is an important branch in statistics and machine learning, aiming to infer causal relationships between variables through observational data. Common causal inference models include path diagrams, potential outcome models, structural equation models, and graphical models.
### Main Contributions of the Paper
1. **Framework Synthesis**: The paper proposes a new framework - causal variational inequalities (CVI) that combines variational inequalities and causal inference models. This framework can handle complex causal relationships in network economics more comprehensively.
2. **Theoretical Basis**: The paper defines in detail the mathematical form of causal variational inequalities and explores their theoretical properties. Specifically, causal variational inequalities are defined as a model \(M = \text{CVI}(F, K)\), where \(F\) is a vector - valued mapping and \(K\) is a closed convex set. Causal interventions are modeled as sub - models \(M_w=\text{CVI}(F_w, K)\), where \(F_w\) reflects the manipulation of certain endogenous variables \(w\).
3. **Application Examples**: The paper shows the application of causal variational inequalities in practical problems through multiple examples, including multi - layer complex network economic non - cooperative games, traffic flow equilibrium problems, etc. These examples not only verify the effectiveness of the framework but also show its wide applicability in different fields.
4. **Algorithm Development**: The paper discusses algorithms for solving causal variational inequalities, including projection algorithms, hypergradient methods, etc. In particular, the paper proposes an incremental two - step projection method, which has high efficiency and stability when dealing with large - scale network economic models.
### Specific Problem Analysis
**Traffic Flow Equilibrium Problem**:
- **Problem Description**: Consider a simple road network from source node 1 to target node 4. Given that 6 vehicles depart from node 1 to node 4 per unit time, how to distribute these vehicles to achieve flow equilibrium?
- **Mathematical Model**: The flow distribution problem can be represented by variational inequalities, with the specific form as follows:
\[
F(x)=\begin{bmatrix}
10&0&0&0&0\\
0&1&0&0&0\\
0&0&1&0&0\\
0&0&0&1&0\\
0&0&0&0&10
\end{bmatrix}
\begin{bmatrix}
x_{12}\\
x_{13}\\
x_{23}\\
x_{24}\\
x_{34}
\end{bmatrix}
+
\begin{bmatrix}
0\\
50\\
10\\
50\\
0
\end{bmatrix}
\]
where \(x_{ij}\) represents the flow from node \(i\) to node \(j\).
- **Causal Intervention**: Consider the causal intervention \(\text{do}(x_{23} = 0)\), that is, vehicles are not allowed to pass through the diagonal edge from node 2 to node 3. In this case, there are only two paths: path \(p_1\)