Variance when only one observation is not NaN

I am trying to calculate a moving variance of the max last 5 years of observations.

I have a tabke that looks like this:

id year A

1 1990 10

1 1991 20

1 1992 20

1 1993 55

1 1994 40

1 1995 65

1 1996 45

2 2001 85

2 2002 NaN

2 2003 90

2 2004 95

2 2005 500

2 2006 550

.

.

.

And I want to calculate this:

id year A pastvarA

1 1990 10 NaN

1 1991 20 NaN

1 1992 20 Var of 10 and 20

1 1993 55 Var of 10, 20 and 20

1 1994 40 Var of 10, 20, 20 and 55

1 1995 65 Var of 10, 20, 20, 55 and 40

1 1996 45 Var of 20, 20, 55, 40 and 65

2 2001 85 NaN

2 2002 NaN NaN

2 2003 90 Var of 85 and NaN = NaN

2 2004 95 Var of 85, NaN and 90 = Var of 85 and 90

2 2005 500 Var of 85, NaN, 90 and 95= Var of 85, 90 and 95

2 2006 550 Var of 85, NaN, 90, 95 and 500 = Var of 85, 90, 95 and 500

.

.

.

I use the code below:

pastvarA=NaN(N,1);

for i=2:5
    for k=i+1:N
        if myTable{k,'id'} == myTable{k-i,'id'} 
            pastvarA(k,1)=var(myTable.A(k-i:k-1),'omitnan'); 
        end
    end
end

I get a result very similar to what I want. The only problem is that I get 0 if there is one observation that is non NaN in the sample that the variance is calculated. But I want to get NaN when that is the case (for example the row: 2 2003 90 Var of 85 and NaN = 0.) the result I get from the code is below:

And I want to calculate this:

id year A pastvarA

1 1990 10 NaN

1 1991 20 NaN

1 1992 20 Var of 10 and 20

1 1993 55 Var of 10, 20 and 20

1 1994 40 Var of 10, 20, 20 and 55

1 1995 65 Var of 10, 20, 20, 55 and 40

1 1996 45 Var of 20, 20, 55, 40 and 65

2 2001 85 NaN

2 2002 NaN NaN

2 2003 90 Var of 85 and NaN = 0

2 2004 95 Var of 85, NaN and 90 = Var of 85 and 90

2 2005 500 Var of 85, NaN, 90 and 95= Var of 85, 90 and 95

2 2006 550 Var of 85, NaN, 90, 95 and 500 = Var of 85, 90, 95 and 500

.

.

.

Answer 1

[Matt J]

A=myTable.A;

pastvarofA=movvar(A,[4,0],'omitnan');
pastNaNs=movsum(~isnan(A),[4,0],'omitnan');

pastvarofA(pastNaNs<=1)=nan;
pastvarofA=[nan;pastvarofA(1:end-1)]

[Mia Dier]

Thank you for your reply Matt! But this code unfortunately, calculates the moving variance without taking the id into account. I want the whole moving variance process to start from each id without countring previous ids' A value in.

[Matt J]

Just generalize the solution with splitapply()

pastVarofA=splitapply(@func,myTable.A,myTable.id);

function pastVarofA = func(A)

    pastvarofA=movvar(A,[4,0],'omitnan');
    pastNaNs=movsum(~isnan(A),[4,0],'omitnan');
    pastvarofA(pastNaNs<=1)=nan;
    pastvarofA=[nan;pastvarofA(1:end-1)];


end

[Mia Dier]

Thank you!