endogenous_prior with missing observation
Hi dynare team,
(Sorry if I posted in the wrong place, I just realised that gitlab seems for developers only)
In v4.6.2, endogenous_prior cannot work with missing observation, but this is not pointed out in the Reference Manual.
I suppose a naive way to fix this is to exclude any period or variables with missing observations in the calculation of statistics.
Currently, I modified the code as follows but not sure if this is statistically correct. Is there a better way to make this work?
Ydemean(:,j)=Y(:,j)-mean(Y(:,j),'omitnan'); Fhat=std(Ydemean,1,'omitnan').^2'; for j=1:n Ydemean(:,j)=Y(:,j)-mean(Y(:,j),'omitnan'); end Fhat=std(Ydemean,1,'omitnan').^2';%Fhat=diag(Ydemean'*Ydemean)/Tsamp; Sgood=find(~any(isnan(Y),2)); Tgood=length(Sgood); hmat=zeros(n,Tgood); % we need ht, where t=1,...,T for t=1:Tgood hmat(:,t)=diag(Ydemean(Sgood(t),:)'*Ydemean(Sgood(t),:))-Fhat; end % To calculate Shat we need C0, C1 and C2 for t=1:Tgood C0=C0+1/Tgood*hmat(:,t)*hmat(:,t)'; end for t=2:Tgood C1=C1+1/(Tgood-1)*hmat(:,t)*hmat(:,t-1)'; end for t=3:Tgood C2=C2+1/(Tgood-2)*hmat(:,t)*hmat(:,t-2)'; end