Consider the growth model with inelastic labor supply, full depreciation, log utility and CRS technology with the Bellman equation be defined as follows: $$V(k)=\max(log(k^\alpha-k')+\beta V(k'))$$ st $$k\geq0\ \text{and}\ \theta k^\alpha-k'\geq0$$

As a guess I have used the usual $V(k)=a+bln(k)$ substituted on the Bellman and have derived $k'=\frac{k^\alpha\beta b}{1+\beta b}$. From this I have found the $k_{ss}=(\frac{1+\beta b}{\beta b})^{1/(\alpha-1)}$ called the non-trivial SS

This is the first SS while the other is $k_{ss}=0$

My question is how can we use the policy function to show that the system converges to the non-trivial steady steady state given any $k_0 > 0$

  • $\begingroup$ A few remarks: your technology is not CRS unless $\alpha = 1$. Also, you forgot to add $\theta$ in your Bellman equation? An issue here is that for $k_t = 0$ (and therefore $k_{t+1} = 0$), your instantaneous utility function is not defined as $\ln(0)$ does not exist. As such, the usual convergence results for the value function of the Bellman equation are quite tricky. In fact the value of $V(0)$ does not exist, so there is no 'optimal policy function' for $k = 0$. $\endgroup$
    – tdm
    May 11, 2021 at 7:09
  • $\begingroup$ @tdm the $\theta$ was confusing me as well that is not included in the Bellman and I wasn't sure if I should have included in the derivation or not. For $k_ss=0$ it makes sense to not have an optimal policy function. But in one of the previous Q. asked it says to derive the policy function $k′(k)$ using the appropriate guess on the value function and to show that there are two SS with one being $k_ss=0$. Maybe the question is not clear? $\endgroup$ May 11, 2021 at 8:06

1 Answer 1


Let's guess that the value function is of the form $a + b \ln(k)$.

Then substituting for $V(k) = a + b \ln(k)$ in the Bellman equation gives: $$ a + b \ln(k) = \max_{k'}\left(\ln(k^\alpha - k') + \beta(a + b \ln(k')\right) $$ The first order condition is given by: $$ \begin{align*} &\frac{-1}{k^\alpha - k'} + \beta b \frac{1}{k'} = 0,\\ \to & k' = \beta b (k^\alpha - k'),\\ \to & k' = \frac{\beta b}{1+ \beta b} k^\alpha \end{align*} $$ If we plug this into the objective function of the Bellman equation, we obtain the following identity: $$ \begin{align*} a + b \ln(k) &= \ln\left(k^\alpha - \frac{\beta b}{1 + \beta b}k^\alpha\right) + \beta\left(a + b \ln\left(\frac{\beta b}{1 + \beta b}k^\alpha\right)\right),\\ &= (\alpha + \beta b \alpha) \ln(k) + \ln\left(1 - \frac{\beta b}{1 + \beta b}\right) + \beta a + \beta b \ln\left(\frac{\beta b}{1 + \beta b}\right) \end{align*} $$ As this holds for all $k (> 0)$ we can equate coefficients on both sides: $$ \begin{align*} a &= \ln\left(\frac{1}{1 + \beta b}\right) + \beta a + \beta b \ln\left(\frac{\beta b}{1 + \beta b}\right),\\ b & = \alpha + \beta b \alpha \end{align*} $$ The second one gives a closed form expression for $b$: $$ b = \frac{\alpha}{1 - \beta \alpha}. $$ Then substituting this into the first order condition gives: $$ \begin{align*} k_{t+1} &= \frac{\beta \frac{\alpha}{1 - \beta \alpha}}{1 + \beta \frac{\alpha}{1 - \beta \alpha}}k_t^\alpha,\\ &= \beta \alpha k^\alpha_t \tag{1} \end{align*} $$

This shows that: $$ k_{t + 1} > k_t \iff \beta \alpha k_t^\alpha > k_t \iff k_t < (\beta \alpha)^{\frac{1}{1 - \alpha}} $$ So the capital stock will rise as long as $k_t$ is below $(\beta \alpha)^{\frac{1}{1 - \alpha}}$ and it will decrease if $k_t$ it is above this threshold.

According to the dynamic equation (1) above, it would appear that $k = 0$ is also a steady state. However, for $k = 0$, the first order conditions are not satisfied and in fact the value function does not exist. Anyway, for $k_t$ very close, it's value will be below $(\beta \alpha)^{\frac{1}{1 - \alpha}}$ so the stock of capital should increase to the unique steady state.

  • $\begingroup$ Thank you so much for answering! $\endgroup$ May 12, 2021 at 18:47

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.