Need some help with a tricky loop

Hi, I'm looking to create a loop. Here's what I have:

SP_returns = 2778x1 array

for i = 2526:2777
  q(i-2525) = quantile(SP_returns(i-2525:i),0.09);
end

Therefore q = 1x252 array

Now I need to express the following as a loop:

tail_SP1 = SP_returns(SP_returns(1:2526) < q(1,1)) - q(1,1);
tail_SP2 = SP_returns(SP_returns(2:2527) < q(1,2)) - q(1,2);
tail_SP3 = SP_returns(SP_returns(3:2528) < q(1,3)) - q(1,3);
...
tail_SP252 = SP_returns(SP_returns(252:2777) < q(1,252)) - 1(1,252);

Basically, tail_SP1 will be in column 1, tail_SP2 will be in column 2 in the array if I do the loop. Problem is, they have different number of observations. For example, tail_SP1 and tail_SP2 may have 227, while tail_SP10 may have 226. The loop won't work. At least not the one I know how to do. I don't think it's even possible to create a matrix which would have different number of values in each column. Perhaps there is another way of doing this thing. I really don't want to have to type 252 lines of code for this. I will appreciate all the help I can get. Thanks.

Answer 1

[Walter Roberson]

Cell arrays can contain vectors of different length.

tail_SP{K} = SP_returns(SP_returns(K:2525+K) < q(1,K)) - q(1,K);

[Julia]

This is what I tried:

for K = 1:252

tail_SP{K} = SP_returns(SP_returns(K:2525+K) < q(1,K)) - q(1,K);

end

The values I'm getting for the first vector are fine, they're all negative and less than q = -0.01, but what I'm getting for the second, third and so on vectors are not. My q is the same for the first few vectors (-0.01) and it's impossible, given my data, to have positive values in tail_SP vectors since (SP_returns < q) give values below -0.01 and if we subtract q from them we will still be getting all negative numbers.

Perhaps I misinterpreted what you were trying to tell me?

[Walter Roberson]

The code I gave replicates what you asked for, but you asked for the wrong thing ;-) Try this:

for K = 1:252

tail_SP{K} = SP_returns(K-1+find(SP_returns(K:2525+K) < q(K))) - q(K);

end

[Julia]

Walter, you're a genius. It worked and it gave me exactly what I wanted. Now, one more thing if you don't mind. The array I'm getting - I've never worked with it before (I'm a new to MATLAB as you may have noticed). This is what I have altogether:

for K = 1:252

tail_SP{K} = SP_returns(K-1+find(SP_returns(K:2525+K) < q(K))) - q(K);

exceedances{K} = -1.*tail_SP{K};

end

I know that I could have just multiplied the top line by -1, but I need the exceedances line to be separate. But that's not what's important.

I now want to use the gpfit function which estimates the parameters of generalized pareto. When I only had one vector for tail_SP I did the following:

paramEsts = gpfit(exceedances);

Now that I have 252 vectors I need a loop which would calculate separate parameters for each vector of exceedances. I have:

for K = 1:252

paramEsts{K} = gpfit(exceedances(K));

end

This doesn't work. So, how can I make it work?

[Walter Roberson]

Very close

for K = 1:252

paramEsts{K} = gpfit(exceedances{K});

end

[Walter Roberson]

By the way: is it possible for any of the tails to be empty? If so then do you want any special handling for that case?

[Julia]

No, they can't be empty.

[Julia]

By the way, the code worked perfectly. Thank you very much. Now I just need to get another 3 loops based on the estimates and I'm done. I'll try to do it myself but something tells me I'll probably be back here in a couple of minutes. :)

[Julia]

Actually, I think I can do the loops, I just need to know how to get a specific value within each vector on its own. I mean every vector has two values. What if I want the first value in the fifth vector? If I type paramEsts{5} I get both values. I tried the variations of it like paramEsts{1,5}. Didn't work.

[Julia]

Never mind, I got it.

[Walter Roberson]

Sorry, it was my turn to cut the front lawn ;-)

Just for completeness, in case someone else reads this and wonders what the solution is, the syntax would be

paramEsts{5}(1)