I am attempting to think through a particular type of game with continuous strategies, with Bayes Nash equilibrium as the solution concept. I first describe the game below, followed by questions.


Consider the following game between two agents: A and B.

  • Agent A draws a cost $k$ from a distribution $F(k)$, which has support on $[0,\overline{k}]$. Knowing this cost, she chooses an action $x\in [0,\infty]$ and thereby incurs a cost $k\cdot x$.

  • The choice of $x$ produces a benefit $R(x)$ for agent B, with $R(0)=0$, $R^\prime(x)>0$ and $R^{\prime\prime}(x)<0$.

  • $R(x)$ is not observed until the game is completed. Instead, A is paid by B for her effort based on a noisy but strictly informative signal $\xi(x)$. Higher $\xi$ always indicates higher $x$ on average.

  • Specifically, B has a prior distribution $\pi (x)$ over $x$ and observes $\xi(x)$. B always pays A her expected product $E(R(x)|\pi,\xi)$ so that B expects a payoff of zero for any value of the signal.

  • Finally, in Bayes Nash equilibrium, we require that $\pi (x)$ is "correct": i.e., given the distribution of costs $F(k)$ and the expected payments as a function of $x$ and $\pi (x)$, actions $x$ actually do have the distribution $\pi(x)$.


  1. There appears to always be an equilibrium to this game with a degenerate prior $\pi(x)$ that has all its mass on zero. In this case, $E(R(x)|\pi,\xi)=0$ for all $x$, and A would never choose $x>0$, confirming the prior. Do you agree?

  2. What are some non-trivial sufficient conditions for there to be additional equilibria in pure strategies (i.e., with $x>0$ in some states of the world)?

  3. How would one characterize these other equilibria if they exist? For example, how would the set of equilibria change if the benefit function increased from $R(x)$ to $\hat{R}(x)$ = $c\cdot R(x)$ where c is a constant greater than one?

Additional Notes

  • I am aware of examples of this kind of game with discrete actions $x$. These are comparatively easy to analyze. What I am struggling with here is how to think about equilibria with continuous actions and therefore also a continuous prior $\pi(x)$. The fixed point problem seems much trickier in this case.

  • EDIT: It seems the way to think about this may be with continuous-state markov chains, though I am not completely sure how to write this problem in that form. Even assuming I am on the right track here, I can so far only find sufficient conditions for the existence of a unique stationary distribution, but no theorems that guarantee non-uniqueness or that suggest a way to do comparative statics on the set of equilibria.

  • Aside from that, feel free to make further assumptions or introduce new notation if it helps.

  • This is the first question I have posted. So constructive feedback is welcome.

  • $\begingroup$ Your question 3. is very unclear. $\endgroup$ – Giskard Jun 8 '17 at 18:42
  • $\begingroup$ Thanks @denesp. I changed it, trying to be more specific. $\endgroup$ – saturno Jun 8 '17 at 18:51
  • 1
    $\begingroup$ Intuitively I don't think there is another equilibrium. Anyways, I would simplify this problem. Pick a functional form for R and a distribution for the signal. Then try to solve it, which should not be too hard for example for a quadratic R and a normal signal. Then generalize, if you think this would yield some sort of interesting insight. $\endgroup$ – Tobias Jun 24 '17 at 18:34

Your Answer

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

Browse other questions tagged or ask your own question.