Problem with OSR

[conglin]

Hi,
I tried to get the optimal values of parameters for simple rules using "osr". However sometimes the parameter values turned to be very very huge around 1000. I guess this is due to the shocks in my model. My model has 10 exogenous shocks, I estimated the standard deviations of these shocks and use these estimates to do simulations.

shocks;
var e_rnf; stderr 0.0636;
var e_finf;stderr 0.0636;
var e_yf;stderr 0.0637;
var e_a;stderr 0.0754;
var e_ix;stderr 2.5744;
var e_efps;stderr 0.5792;
var e_uips;stderr 0.0651;
var e_rn;stderr 0.0638;
var e_pef;stderr 0.3385;
var e_g; stderr 1.1598;
end;

optim_weights;
infcpi 1;
y 1;
end;

osr_params rh_infcpi rh_y rho_rn ;
rh_infcpi =1.8508;
rh_y =0.1225;
rho_rn=0.7539;

osr;

The values I got are:
rh_infcpi 1781.1

rh_y 283.325

rho_rn -921.083

Objective function : 0.947917


But I found if I shut down the shock with the highest standard deviation (var e_ix;stderr 2.5744), the parameter values obtained from "osr" are very good and sensible.

rh_infcpi 1.85777

rh_y 0.387792

rho_rn 0.608402

Objective function : 0.0559598

I don't know what is the reason for this problem. The standard deviation of the shock e_ix is just an estimate of my model, it is much larger than the other ones. Is this a signal that there is something wrong with my estimates? Why osr gives such strange parameter values?

Please help me on this problem. Thanks so much!

Bests
Conglin

[jpfeifer]

Hi,

it is hard to judge what the problem is. Here are some suggestions:

1. Check intermediate values of the shocks. When the OSR coefficients continuously change from the values you describe as sensible to the one you deem to be nonsensical, the problem is most probably not in the OSR-command but in the estimation step.
2. Explain the intuition of the model. What are these shocks? Which tradeoffs are involved? Does it make sense that this one shock is that large? What are the elements in your simple rule? Is it a Taylor rule with interest smoothing and does the interest rate react to output or the output gap? If the latter, how do you define output gap?
3. If you have such a Taylor rule, the one shock may be so large that the policy-maker wants to counteract it as strongly as possible. If it acts simultaneously in the same direction on output and inflation it may even make sense to get such large coefficients. Interest smoothing is completely neglected and the interest rate just extremely reacts to changes in the output gap and inflation.
4. Post the mod-file and the data for people being able to see what happens in your code and look for errors.