The W-footrule coefficient: A copula-based measure of countermonotonicity

This paper introduces the $W$-footrule coefficient, a new copula-based measure of negative association defined as the $L^1$-distance to the countermonotonic copula, establishes its theoretical relationship with Gini's gamma and Spearman's footrule, and provides a statistically rigorous rank-based estimator with proven consistency and asymptotic normality.

The Gist

Imagine you are trying to measure how two things move together. In statistics, we often ask: "Do these two variables dance in sync, or do they move in opposite directions?"

For a long time, statisticians had a great tool called Spearman's Footrule (let's call it the "Positive Dance Meter"). It was excellent at measuring how well two things move together (like two dancers stepping in perfect unison). If they move perfectly together, the meter hits a high score. If they move randomly, it hits zero.

But there was a gap. What if you wanted to measure how well two things move in perfect opposition? Like a seesaw: when one goes up, the other goes down. The old "Positive Dance Meter" could tell you if they were not moving together, but it wasn't designed to specifically measure the quality of that opposition. It treated "moving together" and "moving apart" as just two sides of the same coin, without a dedicated tool for the "moving apart" side.

Enter the new tool: The W-footrule coefficient (Φ).

Think of this paper as the invention of a "Seesaw Quality Meter."

1. The Core Idea: Measuring the "Anti-Diagonal"

Imagine a square grid.

• The diagonal (top-left to bottom-right) represents things moving together (Positive Dependence).
• The anti-diagonal (top-right to bottom-left) represents things moving in perfect opposition (Negative Dependence).

The old meter measured the distance from the diagonal. The new W-footrule measures the distance from the anti-diagonal.

• If the data points hug the anti-diagonal tightly, the score is perfect (indicating perfect negative dependence).
• If they wander away, the score drops.

The authors call this the W-footrule because in the math world, the letter W represents the "Countermonotonic Copula"—the fancy name for that perfect seesaw relationship.

2. The "Gini" Connection: A Balanced Scale

One of the paper's coolest discoveries is a mathematical "aha!" moment. They found that a famous statistic called Gini's Gamma (a general measure of agreement) can be broken down into two parts:

Gini's Gamma = (Positive Dance Meter) + (Seesaw Quality Meter)

It's like saying the total "relationship score" is just the sum of how well they move together plus how well they move apart. This gives us a much clearer picture. If you only looked at the old meter, you might miss the nuance of a strong negative relationship. With this new tool, you can see both sides of the coin clearly.

3. How It Works in Real Life (The Estimator)

The paper doesn't just define the math; it builds a practical tool to calculate this from real data.

• The Method: It uses "ranks." Imagine you have a list of 100 students' heights and weights. You don't care about the exact inches or pounds; you just care that Student A is taller than Student B. The new tool looks at these rankings.
• The Magic: It checks how far the rankings are from the "perfect seesaw" pattern.
• The Result: It produces a number.
  • -1 means they are perfectly opposite (a perfect seesaw).
  • 0 means they are unrelated (random).
  • 0.5 means they are moving perfectly together (the opposite of a seesaw).

4. Why It's Better for "Negative" Relationships

The authors ran thousands of computer simulations (like running a race 10,000 times) to test their new meter against the old one.

• The Finding: When two things are moving in opposite directions (negative dependence), the new W-footrule is much more precise and less "jittery" than the old Spearman's footrule.
• The Analogy: Imagine trying to hear a whisper in a noisy room. The old meter was like a microphone that was good at hearing loud shouts (positive relationships) but got fuzzy when trying to hear the whisper (negative relationships). The new W-footrule is a noise-canceling microphone specifically tuned to pick up that whisper clearly.

5. Robustness: The "Outlier" Proof

In statistics, "robustness" means: "If one weird data point shows up (like a giant in a room of children), does the whole calculation break?"
The paper proves that this new meter is robust. Even if one data point is completely crazy, the meter only wobbles a tiny bit. It won't crash the whole system. This makes it very reliable for real-world data, which is often messy.

Summary

This paper introduces a new statistical ruler designed specifically to measure how well two things move in opposite directions.

• Old way: "Are they moving together? (Yes/No/Sort of)."
• New way: "Are they moving together? AND are they moving apart?"

By giving statisticians a dedicated tool for "negative dependence," this new coefficient helps us understand complex relationships in finance, weather, biology, and social science with much greater clarity, especially when things are moving in opposite directions.

Technical Summary

Here is a detailed technical summary of the paper "The $W$-footrule coefficient: A copula-based measure of countermonotonicity" by Enrique de Amo, David García-Fernandez, and Manuel Úbeda-Flores.

1. Problem Statement

Classical rank-based dependence measures (e.g., Spearman's $\rho$, Kendall's $\tau$) are symmetric; they treat positive and negative dependence on the same scale. While Spearman's footrule ($\varphi_C$) provides a geometric interpretation of deviation from perfect positive dependence (the main diagonal $u=v$) via an $L_1$-distance, there is a lack of a complementary, asymmetric measure specifically designed to quantify deviation from perfect negative dependence (the anti-diagonal $u+v=1$).

The authors aim to construct a new coefficient that:

• Measures the distance of a copula $C$ to the countermonotonic copula $W$ (where $W(u,v) = \max\{u+v-1, 0\}$).
• Serves as a natural counterpart to Spearman's footrule.
• Allows for a decomposition of Gini's gamma ($\gamma_C$) into components representing positive and negative dependence.

2. Methodology

Definition of the $W$-footrule Coefficient ($\Phi_C$)

The authors define the $W$-footrule coefficient, denoted $\Phi_C$, as a normalized $L_1$-distance between the copula $C$ and the countermonotonic copula $W$.

• Initial Formulation: Based on the integral of the distance to the anti-diagonal:
  $$\Phi_C = a \int_0^1 \int_0^1 |1 - u - v| \, dC(u,v) + b$$
• Normalization: The constants $a$ and $b$ are chosen such that:
  • $\Phi_W = -1$ (Perfect negative dependence).
  • $\Phi_M = 1/2$ (Perfect positive dependence, where $M$ is the Fréchet-Hoeffding upper bound).
  • $\Phi_\Pi = 0$ (Independence).
• Closed-Form Representation: Through derivation, the coefficient is expressed as:
  $$\Phi_C = 6 \int_0^1 C(u, 1-u) \, du - 1$$
  Alternatively, using the $L_1$-distance formulation:
  $$\Phi_C = 3 \int_0^1 \int_0^1 |1 - u - v| \, dC(u,v) - 1$$

Theoretical Properties

The paper establishes several key properties of $\Phi_C$:

1. Decomposition of Gini's Gamma: A central theoretical result links Gini's gamma ($\gamma_C$) to Spearman's footrule ($\varphi_C$) and the new coefficient:
   $$\gamma_C = \frac{2}{3} (\varphi_C + \Phi_C)$$
   This interprets Gini's gamma as a balanced combination of deviation from perfect positive dependence and deviation from perfect negative dependence.
2. Bounds and Symmetry: $\Phi_C$ ranges from $-1$ (perfect negative) to $1/2$ (perfect positive). It is invariant under survival copula transformations and exchange of variables.
3. Non-Concordance: Unlike standard concordance measures, $\Phi_C$ is not a measure of concordance because $\Phi_M \neq 1$.

Estimation and Asymptotic Theory

• Estimator: A rank-based estimator $\hat{\Phi}_n$ is proposed using pseudo-observations $\hat{U}_i = R_i/(n+1)$ and $\hat{V}_i = S_i/(n+1)$:
  $$\hat{\Phi}_n = \frac{6}{n} \sum_{i=1}^n (1 - \hat{U}_i - \hat{V}_i)_+ - 1$$
• Consistency: The estimator is proven to be strongly consistent ($\hat{\Phi}_n \xrightarrow{a.s.} \Phi_C$) via the Glivenko-Cantelli theorem for empirical copulas.
• Asymptotic Normality: Under regularity conditions (existence and continuity of first-order partial derivatives of $C$), the estimator is asymptotically normal:
  $$\sqrt{n}(\hat{\Phi}_n - \Phi_C) \xrightarrow{d} N(0, \sigma^2_\Phi)$$
  The asymptotic variance $\sigma^2_\Phi$ is derived using the functional delta method and the influence function of the functional.
• Robustness: The influence function of $\hat{\Phi}_n$ is shown to be uniformly bounded ($|IF| \le 12$), ensuring the estimator is robust in the sense of Hampel (a single outlier cannot drastically change the estimate).

3. Key Results

Analytical Examples

The authors derive closed-form expressions for $\Phi_C$ in specific families:

• Gaussian Copula: For correlation $\theta$, $\Phi_{C_\theta} = \frac{1}{2} + \frac{3}{\pi} \arcsin\left(\frac{\theta-1}{2}\right)$.
• Countermonotonicity: $\Phi_W = -1$.
• Independence: $\Phi_\Pi = 0$.

Simulation Study

A Monte Carlo simulation ($10^4$replications) was conducted for sample sizes$n \in {100, 200, 500}$ across various copula families (Clayton, Gaussian, Gumbel, Frank) with varying dependence strengths.

• Bias and Variance: The estimator $\hat{\Phi}_n$ exhibits small bias that decreases at rate $O(n^{-1})$ and standard deviation decreasing at rate $O(n^{-1/2})$, confirming theoretical predictions.
• Performance under Negative Dependence: The study highlights that $\hat{\Phi}_n$ is significantly more precise (lower bias and comparable or lower variance) than Spearman's footrule estimator ($\hat{\varphi}_n$) when the data exhibits negative dependence.
  • Example: For a Gaussian copula with $\rho = -0.9$ and $n=100$, the bias of $\hat{\Phi}_n$ was $+0.00246$, whereas the bias of $\hat{\varphi}_n$ was $+0.01622$ (more than 6 times larger).
• Boundary Cases: The estimator behaves deterministically for the perfect positive dependence case ($M$), though this lies outside the asymptotic normality assumptions due to lack of density.

4. Significance and Contributions

1. New Measure of Countermonotonicity: The paper introduces $\Phi_C$, filling a gap in the literature by providing a dedicated metric for deviation from perfect negative dependence, complementing existing measures for positive dependence.
2. Theoretical Unification: The decomposition $\gamma_C = \frac{2}{3}(\varphi_C + \Phi_C)$ provides a deeper geometric and statistical understanding of Gini's gamma, separating its sensitivity to positive and negative association.
3. Robust Statistical Inference: The derivation of the asymptotic distribution and the bounded influence function allows for the construction of valid confidence intervals and hypothesis tests (e.g., testing for perfect negative dependence $H_0: \Phi_C = -1$).
4. Practical Utility: The simulation results demonstrate that $\Phi_C$ is the superior tool for analyzing datasets with strong negative correlations, offering higher precision than traditional symmetric measures in that regime.

Conclusion

The $W$-footrule coefficient is a rigorous, rank-based, and robust measure that effectively quantifies countermonotonicity. Its strong theoretical foundation, combined with favorable finite-sample properties, makes it a valuable addition to the toolkit for dependence modeling, particularly in fields where negative association is critical (e.g., risk management, environmental extremes). Future work suggested by the authors includes extensions to higher dimensions and semiparametric efficiency analysis.