by kyri82 » Tue Aug 28, 2012 12:47 pm
Hi,
Dynare stores the simulated variables in a matrix called oo_.endo_simul, with size J-by-N, where J is the number of (all) variables and N is the number of periods you asked. Dynare drops by default the first one-hundred observations in computing statistics to avoid dependence of moments on initial conditions. This can be modified by including the option "drop" following the stoch_simul command. If you are computing statistics using ALL observations of the (simulated) variables, then it is normal they are different. Then, some differences might come from the way you estimate certain moments compared to the way Dynare does, e.g. how exactly correlations or autocorrelations are estimated (I remember I had something like that with autocorrelations). However, these differences should be minor and I think it should be mostly the initial-conditions thing.
Hope that helps. Good luck!!
K.