If $X$ is integrable and $\phi$ is concave, how to show $\phi (X)$ is integrable:
My idea:
define $\phi (X)^{+}=\max \{ \phi (X),0\}$ and $\phi (X)^{-}=\max \{- \phi (X),0\}$
Then $E[\phi (X)]=E[\phi (X)^{+}]+E[\phi (X)^{-}]=E[\max \{ \phi (X),0\}]+E[\max \{- \phi (X),0\}]$
I am not sure whether I can use Jensen on functions of the form $\max \{ X,0\}$
Any help?
This is not true in general. What you actually have is that $\phi(X)_+ \in L_1$ (under the assumptions that $X\in L_1(\mathbb{P})$ and $\phi$ concave).
By changing $\phi$ to $-\phi$, we can work with convex functions instead. We show that if $X\in L_1(\mathbb{P})$, $\phi(X)_-\in L_1(\mathbb{P})$.
Further assume that $\phi$ has domain $(a,b)$ where $-\infty\leq a<b\leq\infty$ and $a<X<b$.
Suppose $x_0=\int X\,d\mathbb{P}$. If $\rho=D^+\phi(x_0)$ then
$$\phi(X)\geq \phi(x_0) + \rho(X-x_0)$$
since $x\mapsto -\min(x,0)=x_-$ is non-increasing and $x_-\leq|x|$, $$ (\phi(X))_-\leq|\phi(x_0)|+|\rho|(|X|+|x_0|) $$
From this, it follows that $\phi(X)_-\in L_1$. $\phi(X)_+$ may not be integrable, however. In any case this is not important for in such case, $\int\phi(X)\,\mathbb{P}$ is well defined and takes the value $+\infty$.