2013年-世界发展银行全球_Integrating_Gravity__The_Role_of_Scale_Invariance_in_Gravity_Models_of_Spatial_Interactions_and_Trade_23页_2mb

Summary of "Integrating Gravity: The Role of Scale Invariance in Gravity Models of Spatial Interactions and Trade"

Core Content

This paper explores the theoretical underpinnings of gravity models in economics, particularly in the context of international trade and spatial interactions. It investigates how different theoretical frameworks can lead to the same empirical gravity equation, emphasizing the role of scale invariance in shaping the functional form of these models.

Main Points

1. Gravity Model Overview

• The gravity model describes bilateral flows (e.g., trade) as a product of the "size" of the origin and destination, and a function of the trade costs (or separation variables).
• The general form of the gravity equation is bi-proportional, meaning it can be expressed as:
  $$
  X_{ij} = A_i B_j K_{ij}
  $$
  where $A_i$ and $B_j$ are size factors, and $K_{ij}$ is the impedance (or trade cost function).

2. Scale Invariance in Gravity Models

• The scale invariance property implies that the model remains unchanged under certain scaling transformations of the size and trade cost parameters.
• This invariance is rooted in the independence of trade costs from the size of the countries, allowing for a consistent representation of flows.
• In the case of exponential trade costs:
  $$
  c_{ij} \rightarrow c_{ij} + \frac{1}{\beta} (\log \lambda_i + \log \kappa_j)
  $$
  This transformation highlights the flexibility of the model in accommodating different size and cost configurations without altering the empirical structure.

3. Theoretical Derivations of the Gravity Equation

• One-dimensional allocation models, such as the CES utility model, derive the gravity equation as an optimal allocation of a representative consumer’s budget.
• Two-dimensional allocation models, like Wilson's, use an entropy-based approach, treating flows as the most probable configuration under budget and market constraints.
• The entropy model is shown to be the only consistent maximum likelihood model that preserves row and column totals, aligning with the bi-proportional structure.

4. Breaking Scale Invariance

• Introducing non-linear interactions between trade costs and flows breaks the scale invariance of the gravity model.
• This is done by modifying the Lagrangian to include a covariance term between trade costs and flows, resulting in a more complex gravity equation:
  $$
  \log X_{ij} = a_i + b_j - \beta \log c_{ij} - \gamma (\log c_{ij} - \langle \log c \rangle)(\log X_{ij} - \langle \log X \rangle)
  $$
• While this extension allows for more flexibility, it does not significantly improve the explanatory power of the model in empirical analysis.

5. Empirical Implementation

• The modified gravity equation is tested on the World Trade matrix, using Poisson regression and OLS.
• The results show that:
  • Poisson regression better fits trade spatial interaction data.
  • The inclusion of the interaction term (i.e., $\gamma$) leads to only marginal improvements in fit (R² or pseudo-R²).
  • The coefficient $\gamma$ is negative and significant, indicating a stronger suppression of small flows for the same trade costs.
  • The predicted values for trade flows remain relatively close across models, suggesting that the modified gravity model does not offer substantial gains over the standard version.

Key Information

• Scale invariance is a defining feature of the standard gravity model, allowing for the separation of size and cost effects.
• The CES utility function is the most common form used in one-dimensional allocation models.
• The entropy model is the only one consistent with the bi-proportional structure and the preservation of marginal totals.
• Non-linear extensions of the gravity model, such as the translog model or modified gravity with covariance terms, can break scale invariance but offer limited empirical benefits.
• The modified gravity model is shown to be a more natural extension that maintains symmetry between origin and destination while introducing non-linearities.

Conclusion

The paper demonstrates that the scale invariance of the gravity model is not merely a mathematical property but has deep implications for the theoretical foundations of trade and spatial interaction models. While non-scale invariant models can be constructed, they do not significantly enhance the empirical performance of the gravity equation. The focus of the paper is on understanding how the structure of the gravity model constrains possible theoretical explanations, rather than on improving its empirical fit.