Dear professor,
after 3 months of work and after purchasing MATLAB (instead of Octave) I have successfully obtained demeaned data, that You suggested I use for random starting points for mode-finding.
I am attaching the obtained charts as printscreens, as well as .MOD file (important: instead of Your .M data file I use .MAT file - is this format okay?).
Is the operation now successful? Last time You wrote:
“In your case, it seems the problem is both about explicit prior boundaries (gamm) and implicit prior truncation due to Blanchard-Kahn violations (the omegas).”
My goal was this:
“First, we try to estimate the DSGE model using a classical maximum likelihood approach to inference. Technically this is done by a Kalman filter approach. To run the Kalman filter procedure, initial values for the parameters of interest must be specified, however. Because likelihoods can have several peaks, we use multiple starting values. We do this to make sure that we indeed estimate the parameters that maximize the likelihood (or, in our case, minimize the negative log-likelihood, which is equivalent). To implement this, we set up a loop that performs 2,500 different draws of vectors of starting values. This takes some time (several hours). Of course, you can reduce the number of draws and therefore the computational time. All this is done by running the script “run.m” placed in the folder “Searching for Starting Values - Full Sample”.”
Model16.mod (3.8 KB)
