EDITED: Variance when only one observation is not NaN

EDITED: I think my problem was not clear from the previous example. So, I added a new one below here:

I am trying to calculate a moving variance of the max last 5 years of observations. If remove 'omitnan' then I get NaN for pastvarofA for all the next 5 observations that follows NaN and I don't want that.

I get 0 if there is an observation that is non NaN in the sample that the variance is calculated. But I want to get NaN when that is the case.

A=[4;NaN;5;5;6;8;9];
pastvarofA=NaN(7,1);

for i=2:5
    for k=i+1:7
        varofA(k,1)=var(A(k-i:k-1), 'omitnan');
    end
end

What I want to get is something like this:

pastvarofA

NaN

NaN

Var of 4 and NaN=NaN

Var of 4, NaN and 5= Var of 4 and 5

Var of 4, NaN, 5 and 5= Var of 4, 5 and 5

Var of 4, NaN, 5, 5 and 6 = Var of 4, 5, 5 and 6

Var of NaN, 5, 5, 6 and 8 = Var of 5, 5, 6 and 8

But what I get is this:

parvarA:

NaN

NaN

Var of 4 and NaN= 0

Var of 4, NaN and 5= Var of 4 and 5

Var of 4, NaN, 5 and 5= Var of 4, 5 and 5

Var of 4, NaN, 5, 5 and 6 = Var of 4, 5, 5 and 6

Var of NaN, 5, 5, 6 and 8 = Var of 5, 5, 6 and 8

[Ive J]

If so, why are you using omitnan then?!

[Mia Dier]

It was probably not clear from previous example why removing 'omitnan' didn't solve the problem. I added a new example above and I hope that the problem is clear now!

Answer 1

[Matt J]

A=[1;2;3;NaN;5;5;6];

varofA  = movvar(A,2,'Endpoints','discard')

varofA = 6×1

0.5000 0.5000 NaN NaN 0 0.5000

●

[Matt J]

For your new example,

A=[4;NaN;5;5;6;8;9];
pastvarofA=movvar(A,[4,0],'omitnan');

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

pastvarofA = 7×1

NaN 0 0 0.5000 0.3333 0.6667 2.0000

●

Answer 2

[Catalytic]

Omit 'omitnan'

[Mia Dier]

It was probably not clear from previous example why removing 'omitnan' didn't solve the problem. I added a new example above and I hope that the problem is clear now!