Consider the following problem, from Constantinides (1990).
\begin{align} V(W_0, x_0) \equiv \max_{c, \alpha} \mathrm{E}_0 \int_0^\infty e^{-\rho s}\gamma^{-1}[c(s) - x(s)]^\gamma \mathrm{d}s, \end{align} subject to $$ \mathrm{d} W(t) =[(\mu-r) \alpha(t)+r] W(t)\mathrm{d} t -c(t) \mathrm{d} t+\sigma \alpha(t) W(t) \mathrm{d} w(t), $$ and $$ \mathrm{d}x(t) = [bc(t) - ax(t)]\mathrm{d}t, $$ where $c$ is consumption, $x$ is the "habit stock" with $x(0) = x_0$ as initial condition, $\alpha$ is the share of wealth $W$ invested in the risky asset with drift $\mu$ and variance $\sigma^2$, and $w(t)$ is a standard Brownian motion.
Theorem 1 in the paper establishes the closed form expressions for consumption, the risky share and the value function, using a standard guess and verify approach. They guess that for some constant $B$ $$ V(W_0, x_0) = B\left(W_0 - \frac{x_0}{r + a - b}\right). $$
My question is: how to guess?
The standard way is to use some kind of homogeneity argument. For example, scaling $W, c, x$ by a constant $\gamma > 0$, we get
$$ V(\gamma W_0, \gamma x_0) = \gamma^{\alpha}V(W_0, x_0), $$ which shows that the value function is homogenous of degree $\alpha$. Then, as one would do for the special case of a standard Merton model ($a = b = 0$), by an appropriate choice of $\gamma$, one gets a guess for the value function, which can then be verified by an application of a verification theorem (see the symmetry section in Chang (2004)). However, since there are two states here, one would need another homogeneity to get something, and I am not seeing anything that can lead me to the guess above. Any suggestions would be greatly appreciated.
