Let $X = (x_*,x^*)$ be an interval in the real line and denote by $\Delta(X)$ the set of simple probability distributions on $X$. Consider a preference relation $\succcurlyeq$ on $\Delta(X)$ that satisfies the axioms of expected utility theory. If $\succcurlyeq$ displays monotonicity with respect to first order stochastic dominance and risk aversion then for all $p \in \Delta(X)$ the certainty equivalent of $p$ exists and is unique.

Sketch of the proof:

1) We define the certainty equivalent of lottery $p$: $CE_p \sim \int_{X}u(x)p(x)dx$

2) We know that if $p$ FOSD $q$ then $p\succcurlyeq q$

3) Because of risk aversion: $\int_{X}u(x)p(x)dx \leq u(\int_{X}xp(x)dx) =u(p)$

4) We want to show that $\exists$ a lottery, call it $s$ such that $CE_p \sim s$; in this respect (from 3 above) we know that $p \succcurlyeq s$; We then pick a particular lottery $q$ such that $p \succcurlyeq q$

5) Since by assumption $\succcurlyeq$ on $\Delta(X)$ satisfies the axioms of expected utility theorem (in particular continuity) there exists a unique $\alpha$ such that $\alpha p + (1-\alpha) q \sim s$ ( see MWG p. 177). We therefore proved that $\exists$ a compound lottery $s$ that is the certainty equivalent of lottery $p$.

QUESTION: Am I missing any detail in the proof?


1 Answer 1


Your notation is a bit misleading: it would be better to write $\mathbb{E}u(p)$ or $U(p)$ for the expected-utility associated with $p$ instead of $u(p)$, and $u(\mathbb{E}p)$ for the utility of the expected value of $p$. Formally $u$ is defined on $X$ and not on $\Delta(X)$.

Regarding your proof, it seems to me that: $(i)$ you don't explain how to find $s$; $(ii)$ you don't find a certainty equivalent, because $s$ is a lottery. What you want to find is a sure monetary prize, i.e. a degenerate lottery, that the decision-maker values equally as the lottery $p$.

For instance, you can notice that, by the monotonicity of $u$, \begin{equation} u(x_{*})\leq \int_{X}{u(x)p(x)dx} \leq u(x^{*}) \end{equation}

In addition, the function $u:x \rightarrow u(x)$ is continuous. Therefore, by the intermediate value theorem, there exists $CE_p \in [x_{*},x^{*}]$ such that $u(CE_p) = \int_{X}{u(x)p(x)dx}$.

For the uniqueness, imagine that $CE'_p$ is another certainty equivalent of $p$, i.e. that $u(CE'_p)=u(CE_p)$. Since $u$ is strictly increasing (which can be seen as a consequence of the monotonicity with respect to first-order stochastic dominance), we obtain $CE'_p=CE_p$.

Notice that you don't need risk aversion to prove the result, but further implies that $CE_p \leq \mathbb{E}p$.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.