I'm reading understanding real business cycle by Plosser.

Here's my crude understanding: For a RBC model, the FOCs of lagrangian together with transversality condition usually forms a nonlinear difference equations system. The standard way of solving this system is by linearization in the neighbourhood of steadystate, solve this problem as a linear difference equations system.

What insight can we get from this solution, except for gaining some knowledge of whether the steadystate is locally stable?


2 Answers 2


Side note: This is one way of solving it - the alternative would be formulating a Bellman equation and iterating on that.

If you assume that the real economy is on or sufficiently close to the steady state, you can also infer about responses to shocks. That is, you can look at the impulse response functions to a change in whatever interests you, and see how the model economy changes. Arguing that we're close enough to that steady state will allow you to see how the economy responds to certain shocks.

Also, in general, you can simulate the economy with (for example TFP) shocks, and look whether the simulated economy looks similar to the real economy. Using this comparison you could judge the model.

This needs arguing that we are close to the steady state - or that convergence happens really fast. Generally, the growth literature around Solow has provided arguments for this.

But your argument is very present in most extensions of the basic RBC model: especially when it is important how close we are to the steady state - when models are more nonlinear. There have been many papers showing that this is the case for standard Neo Keynesian extensions of the RBC model.

  • $\begingroup$ Thank you! But it seems to me, in this regard, using dynamic programming to approximate a global decision rule is strictly better. After all FOC and envelope conditon for Bellman equation implies Euler equations for the original optimization problem. $\endgroup$
    – Epicurus
    Feb 14, 2015 at 16:26
  • $\begingroup$ Big shocks can be problematic with this method even if you are at the steady state. It isn't really enough to assume you are close to the steady state, you also need to assume that the shocks are not so big that the non-linearity of the the problem can be ignored. $\endgroup$
    – BKay
    Feb 14, 2015 at 16:36
  • $\begingroup$ Right, @Bkay, I think that is straightforward. $\endgroup$
    – FooBar
    Feb 14, 2015 at 17:37
  • 1
    $\begingroup$ @Epicurus: Yes, but the more advanced models are too complex to be solved with dynamic programming methods. Be it either due to the "curse of dimensionality" or because non convexities or similar do not allow for standard dynamic programming convergence. $\endgroup$
    – FooBar
    Feb 14, 2015 at 17:56

Higher order approximations such as those generated by Dynare may help a bit in terms of expanding the neighborhood in which the approximation works well, but the fundamental problem remains that the approximation is made about the steady state and deviating too far from the steady state introduces large errors.

Judd, Maliar and Maliar have a paper in Quantitative Economics in 2011 that details a highly nonlinear method that can be used to give pretty good policy function approximations. This method can be finicky with large state spaces, however. So it too suffers from a curse of dimensionality.

Numerically stable and accurate stochastic simulation approaches for solving dynamic economic models KL Judd, L Maliar, S Maliar Quantitative Economics 2 (2), 173-210


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.