correlation and covariance matrix construction

hgeorgako · Post by **hgeorgako** » Wed May 14, 2008 7:47 pm

double post

MadMax · Post by **MadMax** » Wed May 14, 2008 7:53 pm

Lets say you have 1-year daily data (250 data points) for each name, that's 250*85=21250 data points. That's only 21250/85= 5.8 data points per parameter.

The number of data points increase linearly in the number of names, however the number of parameters grows as 2x^2+x in the number of names.

positive definite matrix

MadMax · Post by **MadMax** » Wed May 14, 2008 8:05 pm

hgeorgako,

how about

1- you build up that matrix on moving windows and show them how that matrix is moving around and see if they find any sense in that.

2- construct positions that will look riskless according to that matrix.

rakato · Post by **rakato** » Wed May 14, 2008 8:30 pm

so how many data points per parameter do you think is sufficient?

Chuck · Post by **Chuck** » Thu May 15, 2008 1:46 am

Can someone please tell me where I went wrong? I used Aaron's methodology...

[url=/User%20Files/1540/CORREL%20MATRIX1.xls]Attached File: CORREL MATRIX1.xls[/url]

Tradenator · Post by **Tradenator** » Thu May 15, 2008 2:24 am

I changed your returns from Ln(P2/P1) to P2/P1-1, and removed the 1 day lead. In Aaron's covariance matrix you need to divide by the number of rows. Finally, I removed the *100 you had going from the covariance to correlation. It's also a good idea to check against CORREL() and COVAR() so I did that too. Some of this is just for my calculation preferences, but the division by the sample size in the covariance matrix is important.

[url=/User%20Files/1915/CORREL%20MATRIX1.xls]Attached File: CORREL MATRIX1.xls[/url]

Chuck · Post by **Chuck** » Thu May 15, 2008 7:21 am

Thanks dood! I've always used 'Tools > Data Analysis > Correl > highlight returns' to get the matrix, but seeing how it is actually done using formulas/functions is really helpful. And I forgot to include count(column)...

I'm not doubting you, but why did you use beginning/ending-1 ? It gives me negative returns on up days.

urnash · Post by **urnash** » Thu May 15, 2008 8:51 am

It seems as if there are some tiny mistakes in the matrix named covaraiance using COVAR(). E.g. cells S33, T33, U33, V33 and W33 seems wrong. EDIT: It certainly seems like a nice useful sheet!

Tradenator · Post by **Tradenator** » Thu May 15, 2008 9:16 am

beginning/ending-1

that would be a vertical copulation Blush

Jenian · Post by **Jenian** » Tue May 20, 2008 10:10 am

(empty post)