Including NaN in function for HP-Filter

Hi there, my question is the following: I have a function file to calculate HP-Filtered series of some series y. The problem is the following: If the series y contains some NaN, then the HP-filter function returns a series only with NaN, i.e. up till now it can only produce HP-filtered series when there are no NaN in y. Can somebody adjust my function file such that it works with NaN?

Below the code:

function [cycle,trend]=hpfilter(y,lambda)
long    = size(y,1);
HP=[1+lambda -2*lambda lambda zeros(1,long-3);...
    -2*lambda 1+5*lambda -4*lambda lambda zeros(1,long-4);...
    zeros(long-4,long);...
    zeros(1,long-4) lambda -4*lambda 1+5*lambda -2*lambda;...
    zeros(1,long-3) lambda -2*lambda 1+lambda];
for i=3:long-2;
    HP(i,i-2)=lambda;
    HP(i,i-1)=-4*lambda;
    HP(i,i)=1+6*lambda;
    HP(i,i+1)=-4*lambda;
    HP(i,i+2)=lambda;
end;
trend   = HP\y;
cycle   = y-HP\y;

Thanks very much! Philipp

[David Sanchez]

I tried with

y=str2array('[1 2 e 3]')
y =

       1
       2
     NaN
       3

lambda = .2;

and it seems to work, what's the problem?

Answer 1

[Philipp]

Hello David, maybe I have to be more precise: I run this line of command in order to check for my results:

lambda=1600;
h=log(countrydata.GR.WPU./countrydata.GR.GDPDEF);
x=hpfilter(h,lambda);

later I have everything in a loop which looks like this:

for jj=1:lc;
    eval(['hpw.' Countries{1,jj} '.WPU=hpfilter(log(countrydata.' Countries{1,jj} '.' Var{1,2} './countrydata.' Countries{1,jj} '.' Var{1,9} '),lambda);']);
    eval(['hpw.' Countries{1,jj} '.WPR=hpfilter(log(countrydata.' Countries{1,jj} '.' Var{1,3} './countrydata.' Countries{1,jj} '.' Var{1,9} '),lambda);']);
end

Now h contains some real values and some NaN, but when I execute hpfilter(h,lambda), then x is all NaN?!?

[David Sanchez]

What's the error message you get?

[Philipp]

I dont get an error but each observation is NaN. But what I would like to have is that only the ones that are NaN in h are also NaN in x. Therefor, as mentioned in another forum, the interpolation is not an appropriate solution.