Dynare error for geometric random walk MLE estimation

[gcamilo]

Hi folks, I'm having problems estimating parameters a simple geometric random walk process, I get this error from Dynare:

Code: Select all

??? Attempted to access dr.ghx(1,:); index out of bounds because size(dr.ghx)=[0,0].

Error in ==> kalman_transition_matrix at 40
A(i_n_iv,ic) = dr.ghx(iv,:);

Error in ==> dynare_resolve at 72
[A,B] = kalman_transition_matrix(oo_.dr,iv,ic,M_.exo_nbr);

Error in ==> DsgeLikelihood at 127
[T,R,SteadyState,info] = dynare_resolve('restrict');

Error in ==> initial_estimation_checks at 52
    [fval,cost_flag,ys,trend_coeff,info] =
    DsgeLikelihood(xparam1,gend,data,data_index,number_of_observations,no_more_missing_observations);
   
Error in ==> dynare_estimation_1 at 367
initial_estimation_checks(xparam1,gend,data,data_index,number_of_observations,no_more_missing_observations);

Error in ==> dynare_estimation at 62
    dynare_estimation_1(var_list,varargin{:});

Error in ==> RWmaxlik at 88
dynare_estimation(var_list_);

Error in ==> dynare at 132
evalin('base',fname) ;


My model is very simple:

cgrowth = muc + sigmae*e;

and I'm trying to estimate muc and sigmae. The same data works for another model and I get good estimates. Can any of you help me figure out what's wrong?

[jpfeifer]

I am not exactly sure, but it seems that Dynare cannot handle this case, because the equation you want to estimate is essentially an observation equation, while there is not state transition equation (because there is no endogenous state). Hence, there is no non-degenerate state space representation, which is the reason why the matrix ghx is empty.
Note that you can easily estimate your equation using a standard regression package. Regress cgrowth on constant and you get muc as the constant and sigmae as the variance of the error term.