Stochastic transitivity and quadratic representation functions (Q1320102)

Let \(L\) be the set of finite lotteries with outcomes in \([0,M]\). Let \(\gtrsim\) be a complete, transitive, continuous, monotone (first order stochastic dominance) and strict quasi-concave relation on \(L\). It is assumed that comparing two lotteries \(X\) and \(Y\) the real choice set is the set of compound lotteries \(\{(X,\alpha; Y,1-\alpha)| \alpha\in [0,t]\}\) which can be identified -- assuming the reduction of compound lotteries axiom -- with the set \(\{\alpha X+ (1-\alpha)Y| \alpha\in [0,1]\}\subset L\). If \(V\) is the functional representing \(\gtrsim\) the decision maker will choose a compound lottery \((X,\alpha^*; Y,1- \alpha^*)\), where \(\alpha^*\) is determined by \(\max_ \alpha V(\alpha X+(1- \alpha)Y)\). Let \(C: L\times L\to \mathbb{R}\) be a mapping with \(C(X,Y)= \alpha^*\) if \(X\neq Y\), \(C(X,X)= {1\over 2}\). Then the relation \(\gtrsim\) implies a new relation \(\gtrsim^*\) on \(L\) by defining: \(X\gtrsim^* Y\Leftrightarrow C(X,Y)\gtrsim{1\over 2}\), \(X\sim^* Y\Leftrightarrow C(X,Y)= {1\over 2}\). It is shown: Either the relation \(\gtrsim^*\) is intransitive or the relation \(\gtrsim\) is quadratic (\(\gtrsim\) is quadratic if \(V(x_ 1,p_ 1;\dots; x_ n, p_ n)=\sum^ n_{i=1} \sum^ n_{j=i} p_ i p_ i\varphi(x_ i, x_ j)\) with a symmetric function \(\varphi\)). Since it is optimal for decision makers with a strict quasi-concave preference relation to randomize, the result shows that intransitivity must be observed if the preference relation \(\gtrsim\) is not quadratic. The proofs make strong use of geometrical considerations.