3
$\begingroup$

Consider a 2 player all-pay auction, bidding for $1.

Each submits a bid that is a real number, thus $S_i=[0,\infty)$. The player with a higher bid wins $1, but both players must pay the submitted bid.

Player $i$'s payoff function is:

$v_i(s_i,s_{j})=$

$-s_i$ if $s_i<s_{j}$

($\frac{1}{2}-s_i$) if $s_i=s_{j}$

($1-s_i$) if $s_i>s_j$.

Suppose player $j$ plays a mixed strategy in which she is uniformly choosing a bid between 0 and 1.

I am only interested in the expected payoff from player $i$'s bidding less than 1, because if $s_i$>1, he would win, but also incur a negative payoff.

My question:

Why is Pr$(s_i=s_j)$=0 for any $s_i\in[0,1]$ if player $j$'s mixed strategy is U[0,1]?

My confusion is: the PDF of player $j$ is 1 for any $s_j\in(0,1)$. So if I only consider $s_i<1$ and Pr$(s_i=s_j)$, how do I justify it becomes $0$?

$\endgroup$
5
$\begingroup$

This seems like a basic issue about probability calculus/theory.

The intuition behind the uniform distribution over $[a,b]$ is that all outcomes between and $a$ and $b$ are equally likely. Because of this, one can get an intuitive grasp about the probability that the outcome $x$ falls in a subinterval $[a_1,b_1] \subset [a,b]$. As the ratio of their length is $$ \frac{b_1 - a_1}{b - a}, $$ the subinterval $[a_1,b_1]$ contains $(b_1 - a_1)/(b-a)$ share of all the points in $[a,b]$. Each point is an equally likely outcome, therefore $$ P\left(x \in [a_1,b_1]\right) = \frac{b_1 - a_1}{b - a}. $$ A problem arises if the subinterval is in fact a single point, $a_1$. What is $P\left(x \in \left\{a_1\right\}\right)$?
By the above intuitive reasoning $$ P\left(x \in [a_1,a_1]\right) = \frac{a_1 - a_1}{b - a} = 0. $$ A more rigorous approach:
Let us denote this probability $P\left(x \in \left\{a_1\right\}\right)$ by $\alpha$. As the distribution is uniform, all points are equally likely outcomes and hence for any $a_i \in [a,b]$ we have $$ P\left(x \in \left\{a_i\right\}\right) = \alpha. $$ There are infinitely many (disjoint) points in $[a,b]$. Let $A$ be an arbitrary set of infinitely many such points, that is $$ A = \bigcup_i^{\infty} a_i. $$ Then as probability is sigma additive over disjoint sets $$ P\left(x \in A \right) = P\left(x \in \bigcup_i^{\infty} a_i \right) = \sum_i^{\infty} P\left(x \in \left\{a_i\right\}\right) = \sum_i^{\infty} \alpha. $$ However as $A \subseteq [a,b]$, $P\left(x \in A \right) \leq 1$. But for any $\alpha > 0$ the sum $\sum_i^{\infty} \alpha$ would be infinite. Therefore $\alpha$ cannot be positive, it is zero.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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