What are distortion risk measures, and how do they transform probability to capture risk aversion?

Question

AcadiFi · Accepted Answer

Distortion risk measures transform the probability distribution of losses by applying a distortion (or probability weighting) function to the survival function. This systematically overweights the probability of extreme losses, producing risk measures that reflect how risk-averse agents perceive tail events. **Mathematical Framework:** For a loss random variable X with survival function S(x) = P(X > x), a distortion risk measure is: rho_g(X) = integral from 0 to infinity of g(S(x)) dx where g: [0,1] -> [0,1] is a distortion function satisfying: - g(0) = 0, g(1) = 1 - g is non-decreasing - g is concave (for coherence) ```mermaid graph LR A["Original Survival
Function S(x)"] --> B["Apply Distortion
g(S(x))"] --> C["Distorted
Survival Function"] C --> D["Integrate to get
Risk Measure"] style A fill:#1e3a5f style B fill:#4a2e6e style C fill:#6e2e2e style D fill:#2e5e3e ``` **Common Distortion Functions:** 1. **Proportional Hazard (PH)**: g(u) = u^(1/rho), rho >= 1 Higher rho = more risk aversion. When rho = 1, no distortion (expected value). 2. **Dual Power**: g(u) = 1 - (1-u)^alpha, alpha >= 1 Overweights tail probabilities. Alpha = 1 gives expected value. 3. **Wang Transform**: g(u) = Phi(Phi^{-1}(u) + lambda) Shifts the entire distribution by lambda standard deviations. Used in insurance pricing. **Connection to Known Risk Measures:** - **Expected Value**: g(u) = u (identity, no distortion) - **VaR at alpha**: g(u) = 0 for u <= 1-alpha, g(u) = 1 for u > 1-alpha (step function — NOT concave, hence not coherent) - **ES at alpha**: g(u) = min(u/(1-alpha), 1) (piecewise linear, concave — coherent) **Worked Example (Proportional Hazard, rho = 2):** Camberwell Reinsurance evaluates a catastrophe loss with discrete outcomes: | Loss ($M) | Probability | Survival S(x) | g(S) = sqrt(S) | |---|---|---|---| | 0 | 0.70 | 1.00 | 1.000 | | 5 | 0.15 | 0.30 | 0.548 | | 20 | 0.10 | 0.15 | 0.387 | | 50 | 0.05 | 0.05 | 0.224 | Expected loss = 0.15 x 5 + 0.10 x 20 + 0.05 x 50 = **$5.25M** Distorted risk measure (numerical integration): rho_g = 5 x (1.000 - 0.548) + 15 x (0.548 - 0.387) + 30 x (0.387 - 0.224) + 50 x 0.224 = 5 x 0.452 + 15 x 0.161 + 30 x 0.163 + 11.20 = 2.26 + 2.415 + 4.89 + 11.20 = **$20.77M** (simplified; exact integration differs slightly) The distorted measure ($20.77M) is nearly four times the expected loss ($5.25M), reflecting the heavy overweighting of the $50M catastrophic scenario. **Practical Significance:** Distortion measures are widely used in insurance pricing (known as premium principles). They provide a unified theoretical framework connecting actuarial science, behavioral finance, and quantitative risk management. Explore risk measure theory in our FRM course.

What are distortion risk measures, and how do they transform probability to capture risk aversion?

Master Part I with our FRM Course

Related Questions

Practice Questions