Nov 21, 2016 at 5:04 AM
Edited Jan 3 at 7:58 PM

I am replicating FRBNY model in IRIS and, in order to implement zero lower bound, I need to estimate the timevarying standard deviation of anticipated monetary policy shock in two samples: before and after the financial crisis. Following Jaromir's simple
SPBC tutorial on running Kalman filter with time varying standard deviations of some shocks (more_on_kalman_filter.m), I created J database where I set the anticipated monetary policy shock to zero before 2008q4 and to 0.2 after that. But since this is hard
coded as input to 'vary=' filter option to estimate command, I can not estimate this time varying parameter. Would it be possible for IRIS estimation algorithm to use timevarying dummy so that the standard deviation of some shock is not fixed?
The files are available at
https://github.com/ikarib/FRBNYIRIS
The model requires my modified IRIS fork:
https://iristoolbox.codeplex.com/SourceControl/network/forks/ikarib/IRISdev
Thank you,
Iskander



I found how to solve this problem by setting standard deviation to zero in J database to 'vary=' option for the entire period until 2008q3:
J = struct;
for v=sprintfc('std_rm_sh%d',1:o.nant)
J.(v{1})=tseries(startHist:qq(2008,3),0);
end
filterOpt = {'relative=',false,'objRange=',startHist+2:endHist,'vary=',J};
optimSet = {'MaxFunEvals=',10000,'TolFun=',1e16};
tic
[est,pos,C,H,mest] = estimate(m,d,startHist:endHist,E,'filter=',filterOpt,'optimSet=',optimSet,'sstate=',true,'nosolution=','penalty');
toc

