[Paper Review] Conditional Expectation as Quantile Derivative

Dirk Tasche|ArXiv.org|Apr 19, 2001

Risk and Portfolio Optimization15 references17 citations

TL;DR

This paper establishes that the derivative of a weighted sum's quantile with respect to its weights equals the conditional expectation of the individual random variables given the sum equals the quantile. It further applies this result to derive closed-form expressions for the derivatives of expected shortfall (conditional value-at-risk), demonstrating that expected shortfall is a coherent risk measure under standard assumptions, thereby providing a rigorous foundation for risk capital allocation in finance.

ABSTRACT

For a linear combination of random variables, fix some confidence level and consider the quantile of the combination at this level. We are interested in the partial derivatives of the quantile with respect to the weights of the random variables in the combination. It turns out that under suitable conditions on the joint distribution of the random variables the derivatives exist and coincide with the conditional expectations of the variables given that their combination just equals the quantile. Moreover, using this result, we deduce formulas for the derivatives with respect to the weights of the variables for the so-called expected shortfall (first or higher moments) of the combination. Finally, we study in some more detail the coherence properties of the expected shortfall in case it is defined as a first conditional moment. Key words: quantile; value-at-risk; quantile derivative; conditional expectation; expected shortfall; conditional value-at-risk; coherent risk measure.

Motivation & Objective

• To establish sufficient conditions under which the derivative of a quantile of a weighted sum of random variables with respect to the weights equals the corresponding conditional expectation.
• To extend the quantile derivative result to the expected shortfall (CVaR) risk measure, deriving explicit formulas for its sensitivity to weight changes.
• To rigorously analyze the coherence properties of expected shortfall as a risk measure under standard assumptions in financial risk management.
• To provide a theoretical foundation for capital allocation problems by linking risk measure derivatives to conditional expectations.

Proposed method

• Uses conditional density theory to derive the relationship between quantile derivatives and conditional expectations, assuming the existence of a conditional density for one variable given the others.
• Applies the delta method to express the derivative of the α-quantile of ∑u_j X_j as E[X_i | ∑u_j X_j = Q_α(u)], under regularity conditions on the joint distribution.
• Derives closed-form expressions for the derivative of the expected shortfall risk measure with respect to weights, using the quantile derivative result and properties of conditional expectations.
• Applies Lemma 5.1 to compare conditional expectations over different events, enabling proof of subadditivity and monotonicity for the expected shortfall.
• Establishes that expected shortfall satisfies all four axioms of a coherent risk measure—monotonicity, subadditivity, positive homogeneity, and translation invariance—under the specified assumptions.
• Uses the convex cone structure of portfolios to ensure the validity of conditional expectations and risk measure properties across the domain of interest.

Experimental results

Research questions

• RQ1 Under what conditions does the derivative of the α-quantile of a weighted sum of random variables with respect to the weights equal the conditional expectation of each component given the sum equals the quantile?
• RQ2 How can the quantile derivative result be extended to compute derivatives of the expected shortfall risk measure with respect to portfolio weights?
• RQ3 Is the expected shortfall (CVaR) a coherent risk measure under the standard definition involving conditional expectation over the lower α-tail?
• RQ4 What structural assumptions on the joint distribution of random variables are necessary and sufficient for the quantile derivative to equal the conditional expectation?
• RQ5 How do the coherence properties of expected shortfall depend on the underlying distribution and the existence of conditional densities?

Key findings

• The derivative of the α-quantile of ∑u_j X_j with respect to weight u_i equals E[X_i | ∑u_j X_j = Q_α(u)], under the existence of a conditional density for X_1 given (X_2,…,X_d).
• The derivative of the expected shortfall risk measure with respect to u_i is given by a conditional expectation over the lower α-tail, enabling explicit sensitivity analysis.
• Expected shortfall satisfies all four axioms of a coherent risk measure—monotonicity, subadditivity, positive homogeneity, and translation invariance—when defined as -E[X | X ≤ Q_α(X)].
• The result holds under the assumption that the joint distribution admits a conditional density and that the random variables have finite first moments.
• The coherence of expected shortfall is preserved even when the underlying distribution is not elliptical, provided the technical conditions on conditional densities are met.
• The paper resolves ambiguity in the literature by showing that expected shortfall is coherent under the standard definition involving conditional expectation, despite conflicting claims in earlier works.