A Gini Calibration for Distortion Risk Aversion
TVaR Equivalence
Let \(g:[0,1]\to[0,1]\) be a concave distortion, with \(g(0)=0\) and \(g(1)=1\). For a standard uniform loss \(U\), the distorted price is \[ g(U):=\int_0^1 g(s)ds. \] Write, using \(A\) for area, \[ A_g:=\int_0^1 g(s)ds. \] The TVaR distortion at confidence level \(p\) is \[ t_p(s):=\frac{s}{1-p} \wedge 1. \] For a standard uniform loss, \[ \begin{aligned} \operatorname{TVaR}_p(U)=\int_0^1 t_p(s)ds &=\int_0^{1-p}\frac{s}{1-p}ds+\int_{1-p}^1 1ds \\ &=\frac{1+p}{2}. \end{aligned} \] Therefore the TVaR level that prices \(U\) the same as \(g\) is determined by \[ A_g=\frac{1+p}{2}, \] or \[ p=2A_g-1. \] We write this special value associated with \(g\) as \[ p_g:=2A_p-1. \] The number \(p_g\) is the TVaR-equivalent risk-aversion level of the distortion \(g\). It says: \(g\) has the same average tail-weight concentration as \(\operatorname{TVaR}_{p_g}\).
A Common Scale for Distortion Families
The calibration \(g\mapsto p_g\) gives a useful way to compare different distortion families. Wang, proportional hazards, beta, dual-power, and other concave distortions have native parameters with different meanings that are not directly comparable.
The value \(p_g\) puts them on a common TVaR scale. Instead of saying “use Wang with parameter \(\lambda\)” and “use proportional hazards with parameter \(a\)”, we can say: choose each native parameter so that both distortions have the same \(p_g\). They then price the standard uniform loss the same, and they have the same average concentration of pricing weight in the tail. That does not make the distortions identical. Their marginal weight profiles can still differ materially. But it gives a consistent first-order calibration across families.
The Survival and Quantile Forms
For a nonnegative loss \(X\), the distortion price can be written in survival form as \[ g(X)=\int_0^\infty g\{\mathsf P(X>x)\}dx. \] Equivalently, using quantiles, and assuming enough regularity to avoid distracting atom notation, \[ g(X)=\int_0^1 q_X(1-s)dg(s). \] If \(g\) is differentiable, this becomes \[ g(X)=\int_0^1 q_X(1-s)g'(s)ds. \] Writing \(u=1-s\) gives \[ g(X)=\int_0^1 q_X(u)g'(1-u)du. \] This last form is the key one for the Gini interpretation. It shows that the loss quantile \(q_X(u)\) is weighted by the rank-weight density \(g'(1-u)\). Large losses have \(u\) close to \(1\), so \(1-u\) is close to \(0\). For a concave distortion, \(g'\) is largest near \(0\), so the largest losses receive the largest marginal pricing weights.
Equivalently, using the dual distortion \(\check g(s)=1-g(1-s)\), \[ g(X)=\int_0^1 q_X(u)d\check g(u), \] and, when differentiable, \[ g(X)=\int_0^1 q_X(u)\check g'(u)du. \] The dual distortion \(\check g\) is the bottom-up cumulative version of the same rank-weight allocation. The original concave distortion \(g\) is the top-down cumulative version.
The Gini Interpretation
The usual Lorenz curve is derived bottom-up: order the population from poorest to richest and record \[ L(s)=\text{share of total income earned by the bottom }s\text{ fraction}. \] The usual Gini coefficient is \[ G:=1-2\int_0^1 L(s)ds. \] But we can express the same information top-down. Define \[ T(s)=\text{share of total income earned by the top }s\text{ fraction}. \] Then \[ T(s)=1-L(1-s). \] The top-down curve \(T\) lies above the diagonal, and the same Gini coefficient is \[ G=2\int_0^1 T(s)ds-1. \] This is the form that matches a concave distortion. The distortion \(g(s)\) is the share of total pricing weight (integral of \(g'\)) assigned to the worst \(s\) fraction (from \(0\) to \(s\)) of outcomes. Total pricing weight is \(g(1)=1\), so \(g(s)\) is a genuine cumulative percentage of the total, just as \(T(s)\) is a cumulative percentage of total income.
Thus \(g\) is a top-down Lorenz curve for pricing weights. The equality line \(g(s)=s\) is risk-neutral weighting. A concave distortion lies above the equality line because the worst \(s\) fraction of outcomes receives more than \(s\) of the total pricing weight.
Drawing on this analogy, the coefficient \[ p_g=2\int_0^1 g(s)ds-1= 2A_g -1 \] is therefore a Gini coefficient for the pricing-weight allocation across loss ranks. It is written in the natural top-down orientation for risk.
Risk Ranking of Distortions
The map \[ g\mapsto p_g \] gives a scalar risk ranking of distortions. If \(p_g\) is larger, then \(g\) puts more average pricing weight into the bad tail. Equivalently, \(g\) has the same uniform-loss price as TVaR at a higher confidence level. The monotonic relationship between \(g_p\) and tail risk aversion is a useful feature; native parameters mostly follow the same convention with the exception of the proportional hazard, where tail aversion decreases with \(a\).
The ranking is consistent with pointwise dominance. If \[ g_1(s)\ge g_2(s) \quad\text{for all }s\in[0,1], \] then \[ p_{g_1}\ge p_{g_2}. \] Pointwise dominance is the stronger order. It says that \(g_1\) gives at least as much cumulative weight as \(g_2\) to every worst-\(s\) tail. The scalar \(p_g\) forgets some information by integrating over \(s\). Distortions can cross, and the one with larger \(p_g\) need not dominate pointwise. In that case, \(p_g\) should be read as an average tail-concentration ranking, not as a full ordering of risk aversion at every layer. That loss of information is the price of getting a single number, but the gain is interpretability. The number \(p_g\) has the same units and meaning as a TVaR confidence level. It says which TVaR distortion has the same average concentration of pricing weight as \(g\).
Summary
A concave distortion \(g\) can be read as a top-down cumulative concentration curve for pricing weights. The value \(g(s)\) is the share of total pricing weight assigned to the worst \(s\) fraction of outcomes. The diagonal is risk-neutrality. The excess area above the diagonal measures concentration of attention in the bad tail.
The coefficient \[ p_g = 2A_g =1 = 2\int_0^1 g(s)ds-1 \] has three equivalent interpretations:
- it is the TVaR confidence level that prices a standard uniform loss the same as \(g\);
- it is a common calibration scale for comparing parameters across distortion families;
- it is the top-down Gini coefficient of the pricing-weight allocation over loss ranks.
That is the point of the analogy. We are not borrowing the word “Gini” metaphorically. We are applying the same concentration geometry to the natural top-down object in risk pricing.
References
Acerbi, Carlo, and Dirk Tasche. 2002. “On the Coherence of Expected Shortfall.” Journal of Banking & Finance 26(7): 1487–1503. Acerb2002b?
Gini, Corrado. 1912. Variabilita e Mutabilita. Bologna: Tipografia di Paolo Cuppini. see Ceriani and Verme (2012).
Lorenz, Max O. 1905. “Methods of Measuring the Concentration of Wealth.” Publications of the American Statistical Association 9(70): 209–219. Lorenz (1905)
Wang, Shaun S. 1996. “Premium Calculation by Transforming the Layer Premium Density.” ASTIN Bulletin 26(1): 71–92. Wang (1996)