Edited: Just realised my first post was somewhat misleading and not precise. Thanks to the two commetators that pointed it out.

I am working on an article and ended up wondering for which values of $\gamma>0$ does the following inequality hold:


For $N_i \in [0,1]$ and $\sum_i N_i=1$. This can be rewritten as:

$$\frac{\mathbb{E}_i \left[ c_{i,t}^\gamma\right]}{\mathbb{E}_i \left[c_{i,t}\right]^\gamma}<\frac{\mathbb{E}_i \left[ c_{i,0}^\gamma\right]}{\mathbb{E}_i \left[c_{i,0}\right]^\gamma}$$

Where $c_{i,t}\geq c_{i,0}$ for every $i$. Of course I already noticed that the expresion holds with equality when $\gamma =1$. Somewhat inspired in Jensen´s inequality, my intuition is that it should hold for $\gamma \in (0,1)$ but I haven´t been able to prove it. Any suggestions?

Note: Jensen´s inequality states that $\mathbb{E}\left( \varphi(x) \right) > \varphi \left(\mathbb{E}x \right)$ if $\varphi(x)$ is a convex function.

  • $\begingroup$ If $c_0$ is a known value why do you even bother with expectations on the right hand side? $\endgroup$ – hejseb Jun 9 '16 at 18:08
  • $\begingroup$ hejseb, you were right. Just edited my post so it is not trivial, I was confused about the expectation. $\endgroup$ – user344188 Jun 9 '16 at 19:37

For $0<\gamma <1$, $x^{\gamma}$ is a concave function. Then, by Jensen's inequality \begin{align*} \frac{E(c_t^{\gamma})}{(E(c_t))^{\gamma}} &= E\left(\left(\frac{c_t}{E(c_t)} \right)^{\gamma} \right)\\ &< \left(E\left(\frac{c_t}{E(c_t)}\right) \right)^{\gamma}\\ &=1. \end{align*}

  • $\begingroup$ That works, but I would actually expect $\mathbb{E}(c_t)>c_0$. As it is consumption, which grows over time (sorry for not providing enough context). $\endgroup$ – user344188 Jun 9 '16 at 19:02
  • $\begingroup$ @user344188: I revised. In fact, we do not have to assume that $E(c_t)=c_0$. Note that the right hand side, in your inequality, equals to 1, as $c_0$ is a known quantity. $\endgroup$ – Gordon Jun 9 '16 at 19:10
  • $\begingroup$ You are right. And your answer is in fact correct. But I was confused, $c_0$ shouldn´t be deterministic. Just edited my post in order to provide enough context and correct my mistake. Thanks. If you can check it out that would be great. $\endgroup$ – user344188 Jun 9 '16 at 19:39

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.