Consider three agents $A_i$.

  • $A_1$ moves first and selects $0\leq q_1\in \mathbb{R}$. $A_2$ moves second and selects $0\leq q_2\in \mathbb{R}$. $A_3$ moves third and selects $0\leq p_1\in \mathbb{R}$ and $0\leq p_2\in \mathbb{R}$. $0\leq x\in \mathbb{R}$ is a parameter.
  • $A_1$ tries to maximize $u_1(q_1,p_1, q_2,p_2, x)$, with $\frac{\partial u_1(q_1,p_1, q_2,p_2, x)}{\partial q_1}\geq 0$ and $\frac{\partial u_1(q_1,p_1, q_2,p_2, x)}{\partial p_1}\geq 0$, and independent of $q_2,p_2,x$. $u_1(q_1,p_1, q_2,p_2, x)$ is continuous in all its inputs.
  • $A_2$ tries to maximize $u_2(q_1,p_1, q_2,p_2, x)$, with $\frac{\partial u_2(q_1,p_1, q_2,p_2, x)}{\partial q_2}\geq 0$ and $\frac{\partial u_2(q_1,p_1, q_2,p_2, x)}{\partial p_2}\geq 0$, and independent of $q_1,p_1,x$. $u_2(q_1,p_1, q_2,p_2, x)$ is continuous in all its inputs.
  • $A_3$ tries to maximize $u_3(q_1,p_1, q_2,p_2, x)$, with $u_3(q_1,p_1, q_2,p_2, x)$ increasing for all $p_1<\bar{p}$ and decreasing for all $p_1\geq \bar{p}$; $u_3(q_1,p_1, q_2,p_2, x)$ decreases in $x$ and $q_1$; and $u_3(q_1,p_1, q_2,p_2\rightarrow \infty, x)=-\infty$; $u_3(q_1,p_1, q_2\rightarrow \infty,p_2, x)=-\infty$. $u_3(q_1,p_1, q_2,p_2, x)$ is continuous in all its inputs, but we do not know more about its properties for $p_2,q_2$.
  • We assume that in case that an agent is indifferent between two or more strategies that there exists a decision rule that selects exactly one of these strategies. These rules are known by all agents.

The existence of a unique SPNE is quite clear. However, there could be multiple other equilibria that would lead to the same profit profile (specifically equilibria in which each agent gains the same profits.)

Question: Does $A_3$'s profit decrease in $x$? If yes, why? If not, is there an easy assumption that we need for $u_3(q_1,p_1, q_2,p_2, x)$ behavior with regard to $q_2,p_2$? (I think I miss the forest for the trees) Does the sum of all agents' profit decrease in $x$?

My immediate intuition is yes, because otherwise $A_3$ would be able to pass on all its "losses" to the other agents.

  • $\begingroup$ Can you please carefully double check your question. There are multiple contradictory sentences and typos. Also what is $t_M$ $\endgroup$
    – tdm
    Commented Apr 28, 2021 at 14:51
  • $\begingroup$ I am very sorry for the flod of typos.... I think I got it right this time. Thanks for looking at it. $\endgroup$
    – Paul
    Commented Apr 28, 2021 at 15:24

1 Answer 1


Let's solve by backward induction.

  1. In stage 3, player $A_3$ solves: $$ \max_{p_1, p_2} u_3(p_1, p_2, q_1, q_2, x). $$ Given the assumptions, $A_3$ will set: $p_1 = \overline{p}$ and in general $p_2$ will be a function of the $q_1, q_2$ and $x$, denoted by $p_2(q_1, q_2, x)$. For a solution to exist, you have to assume that $u_3$ is not unbounded in $p_2$.

  2. As $u_2$ is independent of $p_1, q_1, x$, In stage 2, Player $A_2$ solves: $$ \max_{q_2} u_2(q_2, p_2(q_1, q_2, x)). $$ $u_2$ is increasing in $q_2$ and $p_2$. If $p_2(q_1, q_2, x)$ is increasing in $q_2$ then $A_2$ will set $q_2$ as big as possible, so there is no optimal solution. Else, there might be some optimal level of $q_2$, say $q_2(q_1, x)$.

  3. In stage 1, player $A_1$ solves: $$ \max_{q_1} u_1(q_1, \overline{p}). $$ As $u_1$ is increasing in $q_1$ there will probably be no solution to this. So likely this problem is unbounded. If we assume that $q_1 \le \overline{q}_1$ for some fixed value $\overline{q}_1$, then $A_1$ will set $q_1 = \overline{q}_1$.

Going back to the utility of player 3, we therefore see, after substituting: $$ u_3 = u_3(\overline{p}, p_2(\overline{q}_1, q_2(\overline{q}_1, x),x), \overline{q}_1, q_2(\overline{q_1}, x), x). $$ It seems like the change in $u_3$ due to a change in $x$ is ambiguous. As it also influences the optimal choice of $q_2$ (of player $A_2$), and $p_2$ of player $A_3$.

  • $\begingroup$ Thank you very much @tmd. I included two new properties for $u_3$ with regard $p_2$ and $q_2$. I do not yet understand why $A_1$ and $A_2$ who are indifferent to $x$ should essentially take all the losses that arise for $A_3$ when $x$ increases. Do you have the intuition why that might happen? Perhaps I can then add a little bit more structure to rule this behavior out. $\endgroup$
    – Paul
    Commented Apr 28, 2021 at 20:26
  • $\begingroup$ Does the sum of all agents' profit decrease in $x$? $\endgroup$
    – Paul
    Commented Apr 29, 2021 at 10:02

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.