Likelihood computation


I have a question regarding the computation of the likelihood with function dsge_likelihood. I’m estimating the small open economy model in Alvarez-Parra et al. (2013) by ML. I’m using two different data sets for the Mexico. In one estimation I use the first difference of gdp, total consumption and net export-gdp ratio (data 1). In another I use nondurable consumption instead of total consumption (data 2).

For each estimation I calculate the likelihood with dsge_likelihood for both data sets. The problem is that the likelihood is greater for the second estimation for BOTH data sets. I was expecting that the likelihood for each estimations were greater for the data set it uses. Am I using dsge_likelihood correctly?

I’m attaching all the files one needs to run the exercise I just explained. Just run main_AMT.m and look at the variables likelihood_ML_* after running it. The variable likelihood_ML_1 stores the likelihood as computed by dsge_likelihood for data 1 and 2 using the parameters estimated with the data 1. Similarly, likelihood_ML_2 stores the likelihood using the parameters estimated with data 2. As I said above, likelihood_ML_2 (j)> likelihood_ML_1(j) for both j=1,2. Again, I was expecting likelihood_ML_1(1)>likelihood_ML_2(1) and likelihood_ML_1(2)<likelihood_ML_2(2).

Thank you in advance! (30.5 KB)

Comparing likelihoods across different data sets is essentially meaningless.
Do I understand you correctly that a different set of parameters plugged into your model with a particular data set returns a higher likelihood than your original “maximum” likelihood estimate? If yes, you did not find the correct maximum initially. In that case, rerun mode-finding with the second parameter set as the starting point.

Thanks! I guess my initial point is not good enough.