Abstract:Ranking distributions according to a stochastic order has wide applications in diverse areas. Although stochastic dominance has received much attention, convex order, particularly in general dimensions, has yet to be investigated from a statistical point of view. This article addresses this gap by introducing a simple statistical test for convex order based on the Wasserstein projection distance. This projection distance not only encodes whether two distributions are indeed in convex order, but also quantifies the deviation from the desired convex order and produces an optimal convex order approximation. Lipschitz stability of the backward and forward Wasserstein projection distance is proved, which leads to elegant consistency and concentration results of the estimator we employ as our test statistic. Combining these with state of the art results regarding the convergence rate of empirical distributions, we also derive upper bounds for the $p$-value and type I error of our test statistic, as well as upper bounds on the type II error for an appropriate class of strict alternatives. With proper choices of families of distributions, we further attain that the power of the proposed test increases to one as the number of samples grows to infinity. Lastly, we provide an efficient numerical scheme for our test statistic, by way of an entropic Frank-Wolfe algorithm. Experiments based on synthetic data sets illuminate the success of our approach.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is how to effectively count and quantitatively test the convex order relationship between two distributions. Specifically, the authors aim to design a simple yet powerful statistical tool to test the convex order relationship in general dimensions and provide information about the degree of deviation from the expected convex order relationship. In addition, this tool should also be able to find the closest modification so that the two distributions form a convex order relationship.
### Background and Problem Description of the Paper
1. **Importance of Convex Order**
- Convex order is a kind of stochastic order widely used in economics, finance, risk management and other fields. It is used to compare the preference or risk characteristics between two probability distributions.
- For example, in portfolio selection, convex order can be used to evaluate the risks and returns of different investment strategies, helping investors make decisions.
2. **Deficiencies in Existing Research**
- Although other types of stochastic orders such as first - order stochastic dominance have been widely studied, the research on convex order from a statistical perspective is relatively scarce, especially in the multi - dimensional case.
- This gap limits the effective testing of convex order in practical applications, especially in cases where inferences need to be made based on marginal distributions, such as arbitrage detection in financial markets and market equilibrium analysis in labor economics.
### Core Problems of the Paper
- **How to Effectively Test the Convex Order Relationship between Two Distributions?**
- Given two unknown probability measures \(\mu\) and \(\nu\), how to test the hypothesis through sample data \(X_1,\ldots,X_n\sim\mu\) and \(Y_1,\ldots,Y_m\sim\nu\):
\[
H_0:\mu\preceq\nu\quad\text{vs}\quad H_A:\mu\npreceq\nu
\]
- where \(\preceq\) represents the convex order relationship.
- **If They Are Not in a Convex Order Relationship, Then What Is the Degree of Their Deviation from the Expected Convex Order Relationship?**
- This problem is very important for decision - makers because in some application scenarios, allowing a certain range of deviation may be reasonable.
- **If They Are Not in a Convex Order Relationship and One of the Distributions Can Be Changed to Form a Convex Order Relationship, Then What Is the Closest Modification?**
- This is very useful for numerical calculation and simulation because it provides an indirect method to generate distributions that conform to the convex order relationship.
### Solution
To achieve these goals, the authors introduce a statistical testing method based on Wasserstein projection distance. The Wasserstein projection distance can not only encode whether two distributions are indeed in a convex order relationship, but also quantify the degree of deviation and generate the optimal convex order approximation.
#### Key Formulas
- **Wasserstein Projection Distance**
\[
W_2(\mu, P_{\preceq\nu}):=\inf_{\xi\in P_{\preceq\nu}}W_2(\mu,\xi)
\]
\[
W_2(P_{\mu\preceq},\nu):=\inf_{\eta\in P_{\mu\preceq}}W_2(\eta,\nu)
\]
- **Test Statistic**
\[
T_n := W_2(\mu_n, P_{\preceq\nu_m})
\]
- **Hypothesis Testing Rule**
\[
\text{Reject }H_0\text{ if }W_2(\mu_n, P_{\preceq\nu_m})\geq t(\alpha);\text{ otherwise accept}
\]
where \(t(\alpha)\) is a threshold selected according to the significance level \(\alpha\in(0, 1)\) to ensure that the probability of Type I error does not exceed \(\alpha\).
Through this method, the authors not only solve the statistical testing problem of convex order relationship, but also provide means for quantifying deviation and finding the optimal approximation, thus for practical applications.