Matrix Constructed with ||...|| has Too Complicated a Type

[AdamElderfield]

Hi,

I am getting the following error when trying to create a matrix which I can pass to the DLM instruction:

## SX26. Matrix Constructed with ||...|| has Too Complicated a Type
>>>> ||<<<<

The matrix has one parameter estimate which is a wrapped up in a function (similar to the Nelson-Siegal model), the vector tau has some numbers in it which I want to index:

Code: Select all

dec vect tau_f(7) 
dec vect tau_s(7) 
dec real tenor

dofor tenor = 120 3 12 12 24 12 24
 compute tau_f(%doforpass) = tenor
end dofor tenor

dofor tenor = 0 0 0 0 12 0 12
 compute tau_s(%doforpass) = tenor
end dofor tenor



frml Cf = ||1.0,-((exp(-lambda*tau_s(1))-exp(-lambda*tau_f(1)))/(tau_f(1)-tau_s(1))),((tau_s(1)*exp(-lambda*tau_s(1))-tau_f(1)*exp(-lambda*tau_f(1)))/(tau_f(1)-tau_s(1)))+((exp(-lambda*tau_s(1))-exp(-lambda*tau_f(1)))/(tau_f(1)-tau_s(1))),0.0,0.0,0.0|$
          ||1.0,-((exp(-lambda*tau_s(2))-exp(-lambda*tau_f(2)))/(tau_f(2)-tau_s(2))),((tau_s(2)*exp(-lambda*tau_s(2))-tau_f(2)*exp(-lambda*tau_f(2)))/(tau_f(2)-tau_s(2)))+((exp(-lambda*tau_s(2))-exp(-lambda*tau_f(2)))/(tau_f(2)-tau_s(2))),0.0,0.0,0.0|$
          ||1.0,-((exp(-lambda*tau_s(3))-exp(-lambda*tau_f(3)))/(tau_f(3)-tau_s(3))),((tau_s(3)*exp(-lambda*tau_s(3))-tau_f(3)*exp(-lambda*tau_f(3)))/(tau_f(3)-tau_s(3)))+((exp(-lambda*tau_s(3))-exp(-lambda*tau_f(3)))/(tau_f(3)-tau_s(3))),0.0,0.0,0.0|$
          ||1.0,-((exp(-lambda*tau_s(4))-exp(-lambda*tau_f(4)))/(tau_f(4)-tau_s(4))),((tau_s(4)*exp(-lambda*tau_s(4))-tau_f(4)*exp(-lambda*tau_f(4)))/(tau_f(4)-tau_s(4)))+((exp(-lambda*tau_s(4))-exp(-lambda*tau_f(4)))/(tau_f(4)-tau_s(4))),0.0,0.0,0.0|$
          ||1.0,-((exp(-lambda*tau_s(5))-exp(-lambda*tau_f(5)))/(tau_f(5)-tau_s(5))),((tau_s(5)*exp(-lambda*tau_s(5))-tau_f(5)*exp(-lambda*tau_f(5)))/(tau_f(5)-tau_s(5)))+((exp(-lambda*tau_s(5))-exp(-lambda*tau_f(5)))/(tau_f(5)-tau_s(5))),0.0,0.0,0.0|$
          ||1.0,-((exp(-lambda*tau_s(6))-exp(-lambda*tau_f(6)))/(tau_f(6)-tau_s(6))),((tau_s(6)*exp(-lambda*tau_s(6))-tau_f(6)*exp(-lambda*tau_f(6)))/(tau_f(6)-tau_s(6)))+((exp(-lambda*tau_s(6))-exp(-lambda*tau_f(6)))/(tau_f(6)-tau_s(6))),0.0,0.0,0.0|$
          ||1.0,-((exp(-lambda*tau_s(7))-exp(-lambda*tau_f(7)))/(tau_f(7)-tau_s(7))),((tau_s(7)*exp(-lambda*tau_s(7))-tau_f(7)*exp(-lambda*tau_f(7)))/(tau_f(7)-tau_s(7)))+((exp(-lambda*tau_s(7))-exp(-lambda*tau_f(7)))/(tau_f(7)-tau_s(7))),0.0,0.0,0.0||

Do I need to hard enter the tau values?

[TomDoan]

The problem you have is that you keep repeating the ||'s at the start of the 2nd and later lines, which adds another layer of embedded matrices. (It's trying to build a RECT[RECT[RECT[RECT[....] until it hits the nesting limit.)

It's usually better to use a FUNCTION rather than a FRML if the FRML starts to get unreadably complicated. In this case, since each row is basically the same except for the row number, the FUNCTION to compute it is fairly simple. If I'm reading this correctly, you want:

Code: Select all

dec rect cf0(7,6)
ewise cf0(i,j)=(j==1)
function cfx t
type integer t
do i=1,7
  compute cf0(i,2)=-((exp(-lambda*tau_s(i))-exp(-lambda*tau_f(i)))/(tau_f(i)-tau_s(i)))
  compute cf0(i,3)=((tau_s(i)*exp(-lambda*tau_s(i))-tau_f(i)*exp(-lambda*tau_f(i)))/(tau_f(i)-tau_s(i)))+((exp(-lambda*tau_s(i))-exp(-lambda*tau_f(i)))/(tau_f(i)-tau_s(i)))
end do i
compute cfx=cf0
end

You can then use C=CFX(T) on the DLM instruction.

[AdamElderfield]

Thanks for the tip Tom, I will try that in future!

[AdamElderfield]

Hi Tom,

I've tried your function, thanks. I am getting the following error:

## SX22. Expected Type REAL, Got RECTANGULAR[REAL] Instead
>>>> compute cfx=cf0<<<<

I'm not sure how to proceed?

Thanks

Adam