topic Re: PROC IML to solve non-linear equation in SAS/IML Software and Matrix Computations

PROC IML to solve non-linear equation

MAC1430 — Tue, 07 Sep 2021 04:43:44 GMT

Hi everyone,

I need to calculate the unknown discount rate r in the equation below, all the other variables in the model are known.

Anyone knows how to solve the equation below using PROC IML, Froot function please?

A data sample looks like below:

data test;
input stock$     y	B	P	F	g;
datalines;
A	1	12.1492	80.16	3.8886	0.3038
A	2	15.3785	63.54	4.8659	0.3769
A	3	11.5507	41.19	4.5454	0.205
A	4	15.1777	35.28	4.3858	0.1478
A	5	14.9545	52	5.7079	0.0644
A	6	15.8837	50.78	6.1861	-0.0636
A	7	19.3065	67.64	6.7226	-0.0915
A	8	21.6415	72.05	7.7686	-0.0553
A	9	23.6561	74.51	8.804	0.0005
A	10	27.2373	43.05	9.0739	0.123
B	1	6.3668	37.8	2.7893	0.3038
B	2	7.6967	35.4	3.238	0.3769
B	3	8.6812	29.3	3.2122	0.205
B	4	8.999	30.9	3.2352	0.1478
B	5	9.5064	29.75	3.5381	0.0644
B	6	10.1293	39.3	3.4308	-0.0636
B	7	12.2344	43.3	3.7561	-0.0915
B	8	13.2885	52	4.2523	-0.0553
B	9	13.9864	41.6	4.7464	0.0005
B	10	14.043	50.2	4.2598	0.123
C	1	1.4639	24.19	0.3939	0.3038
C	2	2.3384	19.52	1.3403	0.3769
C	3	2.626	7.96	-1.0251	0.205
C	4	2.7283	15.72	0.3156	0.1478
C	5	2.3651	11.81	0.049	0.0644
C	6	2.8771	16.9	0.964	-0.0636
C	7	3.532	18.84	1.1485	-0.0915
C	8	4.5199	21.66	2.0169	-0.0553
C	9	4.379	12.75	2.0582	0.0005
C	10	4.6028	24	0.9412	0.123
;

If you know the solution to the problem, please comment, thank you very much.

Re: PROC IML to solve non-linear equation

MAC1430 — Tue, 07 Sep 2021 23:31:34 GMT

Another member has already helped me finding the solution using Proc fcmp.
https://communities.sas.com/t5/SAS-Programming/Solving-Equation/m-p/766280#M242843
I am wondering if someone can help me to find solution using Proc IML. Thank you very much.

Re: PROC IML to solve non-linear equation

Rick_SAS — Tue, 07 Sep 2021 10:15:54 GMT

That is a strange formula. Usually time series formulas use "lags," but yours has "leads." You have to tell us how you intend to evaluate the equation at the end of the time series. Your example data has y=1:10. Tell us how you want to evaluate expressions such as F_{y+tau} when y=10, for tau=1:5. When I look at your formula for y=10, I see

p_10 = B_10 + (TERMS_I_CANT_EVALUATE)

Do you only want solutions for y<=5?

Re: PROC IML to solve non-linear equation

MAC1430 — Tue, 07 Sep 2021 23:32:24 GMT

Hi Rick,

Thank you very much.

Yes, solution for y<=5 would be great, assuming the first 5 year are known data, last 5 year are forecast data.

Please let me know if you need more information, thank you for the help.

Re: PROC IML to solve non-linear equation

Rick_SAS — Wed, 08 Sep 2021 09:55:44 GMT

In the formula, the rate, r, does not have a subscript, so the equations imply that there is a single value of r that solves these equations simultaneously for all values of the time index, y. However, in the previous thread, you marked a correct solution, and that program computes a different rate for each value of the time index. Is that really what you want? Or are you trying to find a value of r that solves the equations in a least-squares sense?

So, I guess I am confused by whether the equation is supposed to be solved FOR EACH index, y, or for ALL indices simultaneously, which requires finding a least-squares solution.

I am also wondering why you want to solve this problem by using SAS/IML, if you already have a solution by using the DATA step? What are you hoping to gain from a second solution?

Re: PROC IML to solve non-linear equation

MAC1430 — Thu, 09 Sep 2021 01:03:45 GMT

Thank you very much for the comment.

In the equation, p, denote the stock price; r, denote a discount rate; all the vars (p,B,F,g,r) are specific to each stock.

By solving the equation, I want to estimate the values of r, for each stock and year (y).

I want a solution using proc IML because it is advised that it’s most suitable to solve problems like this. I want to learn how to use porc IML for this equation, with the hope to solve other problems like this. It’s a learning process for me, and hope the solution can benefit other SAS users as well.

Please feel free to comment if more info needed, thank you for the help.