Conditional Value at Risk vs Value at Risk in Non-Normal Equity Distributions

The Structural Blind Spot of Value at Risk

Value at Risk (VaR) remains the most common metric for institutional portfolio risk reporting. Defined as the maximum expected loss over a specific time horizon at a given confidence level $\alpha$, VaR is expressed as:

$P(L > \text{VaR}_\alpha) = 1 - \alpha$

While intuitive, VaR contains a critical structural blind spot: it functions strictly as a boundary quantile cutoff. It provides no information regarding the severity of losses once that threshold is breached. In a benign Gaussian distribution, tail decay is exponential. However, realized equity returns exhibit heavy tails, negative skewness, and extreme kurtosis. Two distinct portfolios can exhibit identical 99% 1-day VaR values, yet one may incur a 3% excess loss in the tail while the other suffers a catastrophic 35% liquidation drawdown.

Coherent Risk Measures and the Sub-Additivity Failure

In their seminal paper, Artzner et al. (1999) defined four mathematical axioms that any coherent risk measure $\rho$ must satisfy: monotonicity, sub-additivity, positive homogeneity, and translation invariance.

VaR fails the fundamental axiom of sub-additivity:

$\rho(X + Y) \le \rho(X) + \rho(Y)$

Sub-additivity dictates that combining positions must not generate more risk than the sum of their individual risks. When asset returns are non-normal or contain asymmetric tail risk, the VaR of a diversified portfolio can exceed the sum of individual constituent VaRs. Consequently, optimizing a portfolio on VaR can actively penalize true diversification and reward strategies that conceal risk in catastrophic, unmeasured tail outcomes.

Expected Shortfall and Tail Convexity

Conditional Value at Risk (CVaR), also known as Expected Shortfall, resolves this failure by calculating the expected loss conditional on exceeding the VaR threshold:

$\text{CVaR}_\alpha(L) = \mathbb{E}[L \mid L \ge \text{VaR}_\alpha(L)]$

As proven by Rockafellar and Uryasev (2000), CVaR satisfies all coherent risk axioms. Because CVaR is convex and continuous, it can be optimized efficiently through linear programming rather than non-convex combinatorial methods. This mathematical rigor led the Basel Committee on Banking Supervision (BCBS 457) to transition regulatory market risk standards from 99% VaR to 97.5% Expected Shortfall.

Allocators utilizing the Parallax Risk Management Framework and enterprise platform analytics apply CVaR-based risk budgeting to ensure portfolio tail exposures are fully quantified under non-normal market regimes.