Separating marginal utility and probabilistic risk aversion (Q1315454)

The famous RDU-model developed by Quiggin is characterized and risk aversion within the RDU-model is analyzed. Theorem 12 gives the following characterization: A preference relation \(\gtrsim\) can be represented by an RDU-model, i.e. by a utility function \(U\) and a probability transformation \(\varphi\) with (for the special case of finite lotteries) \[ \text{RDU}(p_ 1,x_ 1;\dots; p_ n, x_ n)= \sum^ n_{i=1} \varphi(p_ 1+\cdots+ p_ i)- \varphi(p_ 1+\cdots+ p_{i-1}) U(x_ i), \] if and only if \(\gtrsim\) is a weak order with (1) for every probability distribution exists a certainly equivalent, (2) a weak continuity assumption (simple continuity), (3) stochastic dominance, (4) tradeoff consistency (with respect to rank-ordered outcomes). The last condition was called `absence of contradictory tradeoffs' in earlier papers of the author: Consider four lotteries 1, 2, 3, 4 where lotteries 1, 2 and lotteries 3, 4 have equal probabilities respectively. Denote by \(1\alpha\) lottery 1 where e.g. the smallest outcome is replaced by \(\alpha\) and define \(1\gamma\), \(2\beta\), \(2\delta\), \(3\alpha\), \(3\gamma\), \(4\beta\), \(4\delta\) similarly. Tradeoff consistency implies that \(1\alpha\gtrsim 2\beta\), \(1\gamma\lesssim 2\delta\), \(3\gamma\gtrsim 4\delta\), \(3\alpha< 4\beta\), is not possible for all outcomes \(\alpha\), \(\beta\), \(\gamma\), \(\delta\). In part 4 the author characterizes marginal utility on the one hand and probabilistic risk aversion independently of marginal utility on the other hand. If \(\gtrsim^ 1\) and \(\gtrsim^ 2\) are represented by RDU's \(\varphi^ 1\), \(U^ 1\) and \(\varphi^ 2 U^ 2\), respectively, then \(\gtrsim^ 2\) exhibits a stronger decrease of marginal utility than \(\gtrsim^ 1\) (a high-payment tradeoff cannot come out better with \(\gtrsim^ 2\) than with \(\gtrsim^ 1\)) if and only if \(U^ 2\) is a concave transformation of \(U^ 1: U^ 2= \varphi\circ U^ 1\) (with \(\varphi\) convex, continuous and increasing). Moreover, the author shows that aversion to probabilistic risk within RDU-models -- interpreted as quasi-convexity with respect to probability mixtures \((P\gtrsim Q\Rightarrow P\gtrsim \lambda P+ (1-\lambda)Q)\) -- is equivalent to convexity of \(\varphi\). The paper contains more important results (e.g. a characterization of EU by the attitude with respect to probabilistic risk) and an interesting discussion of utility and risk attitude in the past.