Large variance of disturbance

Dear Prof Johannes,

I am doubting on my estimation result as follows:

I estimate the investment-specific process by writing:

K / Z = K(-1) / Z + I*(…)

where Z = rhoZ*Z(-1) + eps_Z.

I suppose the prior eps_Z ~ N(0.03,0.1^2) and estimate this with other params. While everything is good, but estimated of eps_Z is very large (~1.2-1.4). Should I concern about it?

I do think Z is fluctuating around 1 +/- eps_z, so if se(eps_z) is large it suggests investment quality subjects to high volatility.

I wonder could my estimation be wrong? Is it possible that iid innovation processes have large variance?

Thanks in advance,

Hi, the equation you give looks weird:

  • First, you divide K by Z which is supposed to be zero mean,
  • Second, you divide K_t by Z_t and K_{t-1} by Z_t (same timing for Z,
  • Third, your shock Z does not look like an investment specific shock. Do you have a reference for specifying Z this way?


Hi Prof Stephan,

Sorry for my confusing writing.
I mean Z(t) is investment propagation process as Zbar = 1.
My capital motion is:

K(t) = (1-delta)*K(t-1) + Z(t)I(t)(1 - S(t)), where S(t) is adjustment-costs function.
Hence, Z(t) is investment-specific shocks (as I understand), which is log(Z(t)) = rho_Z
log(Z(t-1)) + eps_Z(t).

In my model, all shocks are described as above and I can estimate all eps_xx reasonably excepting for investment specific shock.

Could you please give me an idea? Thank you very much.