Counterexamples to "Transitive Regret"

Abstract

Theorem 1 in Bikhchandani & Segal (2011) suggests that a complete, transitive, monotonic, and continuous preference is regret based if and only if it is expected utility. Their Proposition 1 suggests that transitivity and continuity of a regret-based preference implies an equivalence condition: if random variables $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}X$ and $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}Y$ have the same distribution, then $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}X\sim Y$. We give counterexamples to Proposition 1.

1 Notations

Let $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}L$ be the set of real finite-valued random variables over $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}(S,\Sigma,P)$ with $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}S=[0,1]$, $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}\Sigma$ being the Borel $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}\sigma$-algebra, $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}P$ being the Lebesgue measure, and the set of outcomes being the bounded interval $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}[\underline{x},\bar{x}]$. The decision maker has a preference relation $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}\succsim$ over $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}L$. Denote events by $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}S_{i}$.

Definition 1. The continuous function $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}\psi:[\underline{x},\bar{x}]\times[\underline{x},\bar{x}]\rightarrow\mathbb{R}$ is a regret function if for all $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}x$, $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}\psi(x,x)=0$, $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}\psi(x,y)$ is strictly increasing in $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}x$ and strictly decreasing in $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}y$.

If in some event $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}X$ yields $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}x$ and $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}Y$ yields $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}y$, then $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}\psi(x,y)$ is a measure of the decision maker’s ex post feelings about the choice of $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}X$ over $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}Y$. This leads to the next definition.

Definition 2. Let $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}X,Y\in L$, where $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}X=(x_{1},S_{1};\ldots;x_{n},S_{n})$ and $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}Y=(y_{1},S_{1};\ldots;y_{n},S_{n})$. The regret lottery evaluating the choice of $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}X$ over $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}Y$ is

$$\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}\Psi(X,Y)=(\psi(x_{1},y_{1}),p_{1};\ldots;\psi(x_{n},y_{n}),p_{n}),$$

where $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}p_{i}=P(S_{i})$, $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}i=1,\ldots,n$.

Refer to $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}\psi$ and $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}\Psi$ as a regret function and a regret lottery. The main model follows.

Definition 3. The preference relation $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}\succsim$ is regret based if there exists a regret function $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}\psi$ and a continuous functional $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}V$ that is defined over regret lotteries such that for any $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}X,Y\in L$,

$$\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}X\succsim Y\text{ if and only if }V(\Psi(X,Y))\geq 0.$$

2 Counterexamples

Proposition 1 (Bikhchandani & Segal 2011).

Let $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}\succsim$ be transitive, continuous, and regret based. For all $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}X$ and $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}Y$, if they have the same cumulative distribution with respect to $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}P$, then $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}X\sim Y$.

One difference between Theorem 1 and Proposition 1 in Bikhchandani& Segal (2011) is that Proposition 1 drops two axioms: monotonicity and completeness. The following counterexamples imply that the assumptions for Proposition 1 is too weak due to the lack of completeness axiom.

Assume $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}\psi(x,y)=x-y$, a continuous regret function.

Counterexample 1: Under the constant representation $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}V(\Psi(X,Y))=-1$, $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}X\not\succsim Y$ if $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}X,Y$ have the same distribution.

It will be proven that $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}V$ satisfies transitivity, continuity, and regret-based.

Proof.

First, observe $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}V(\Psi(X,Y))=-1$ for any regret lotteries. $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}V$ is regret-based because a constant function is continuous and $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}\psi(x,y)=x-y$ is a continuous regret function.

Transitivity will be proven by contradiction. Assume transitivity does not hold; then, there must exist $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}X,Y,Z$ such that $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}X\succsim Y$, $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}Y\succsim Z$, and $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}X\not\succsim Z$. However, $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}V(\Psi(X,Y))<0$ implies $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}X\not\succsim Y$, a contradiction! So transitivity holds.

Since for all $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}X,Y:$ $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}V\not\geq 0$, we have $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}X\not\succsim Y$ for any $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}X,Y$. The upper contour sets and lower contour sets of $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}\succsim$ are all empty sets and thus closed sets, satisfying continuity. ∎

Counterexample 2: Under the representation $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}V(\Psi(X,Y))=-\sum_{i}|\psi(x_{i},y_{i})p_{i}|$, $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}X\not\succsim Y$ if $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}X,Y$ have the same distribution and $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}X\neq Y$.

Proof.

First, observe that $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}-\sum_{i}|\psi(x_{i},y_{i})p_{i}|$ is regret-based because the sum of absolute value functions is continuous and $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}\psi(x,y)=x-y$ is a continuous regret function.

Assume $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}X\succsim Y$ and $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}Y\succsim Z$. Observe that $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}-\sum_{i}|\psi(x_{i},y_{i})p_{i}|\leq 0$ and the equality holds if $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}\psi(x_{i},y_{i})=0$ for all $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}i$. Since $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}X\succsim Y$ implies $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}V(\Psi(X,Y))\geq 0$, we have $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}V(\Psi(X,Y))=0$. Thus, $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}x_{i}=y_{i}$ for all $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}i$ and $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}X=Y$. Similarly, $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}Y=Z$. Therefore, $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}X\succsim Z$. Transitivity holds.

Observe that the upper contour set $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}\{X:X\succsim Y\}=\{X\}$ is a singleton and that the lower contour set is also a singleton. Therefore, the upper contour sets and the lower contour sets of $\color[rgb]{0,0,1}\definecolor[named]{pgfstrokecolor}{rgb}{0,0,1}\succsim$ are singleton sets and therefore closed sets, satisfying continuity. ∎

Proposition 1 is the first step in Bikhchandani& Segal (2011)’s proof (page 98) for Theorem 1, and is applied in their Section 3.3, page 100.