The RATS Software Forum

Posted: **Tue Feb 02, 2016 4:51 pm**

This code

calendar(q) 1969 1

compute begin_samp = 1969:1
compute end_samp   = 2006:4
compute max_lags = 4

open data "sandbox_data.xls"
    data(format = xls, org = columns)
close data

system(model = mdl)
    variables lgdpk dlp
    lags 1 to max_lags
    deterministic constant
end(system)
estimate(model = mdl)

compute prior_coeffs = %modelgetcoeffs(mdl)
compute post_coeffs = prior_coeffs

* trivial example: zero out intercept terms
do i = 1, 2
    compute post_coeffs(%rows(post_coeffs), i) = 0
end do i

compute %modelsetcoeffs(mdl, post_coeffs)
compute eq = %modeleqn(mdl, 1)
linreg(create, equation = eq)

displays this output:

Code: Select all

Linear Regression - Estimation by User
Dependent Variable LGDPK
Quarterly Data From 1970:02 To 2006:04
Usable Observations                       147
Degrees of Freedom                        138
Mean of Dependent Variable       8.7984421340
Std Error of Dependent Variable  0.3252359768
Standard Error of Estimate       9.0869764659
Sum of Squared Residuals         11395.093498
Durbin-Watson Statistic           1.6276e-006

    Variable                        Coeff      Std Error      T-Stat      Signif
************************************************************************************
1.  LGDPK{1}                         0.000000     0.000000      0.00000  0.00000000
2.  LGDPK{2}                         0.000000     0.000000      0.00000  0.00000000
3.  LGDPK{3}                         0.000000     0.000000      0.00000  0.00000000
4.  LGDPK{4}                         0.000000     0.000000      0.00000  0.00000000
5.  DLP{1}                           0.000000     0.000000      0.00000  0.00000000
6.  DLP{2}                           0.000000     0.000000      0.00000  0.00000000
7.  DLP{3}                           0.000000     0.000000      0.00000  0.00000000
8.  DLP{4}                           0.000000     0.000000      0.00000  0.00000000
9.  Constant                         0.000000     0.000000      0.00000  0.00000000

I checked the new model that %modelsetcoeffs returns, and the coefficients are correct (in that the model includes the updated coefficients), but why doesn't LINREG set the coefficients correctly? If I run

Code: Select all

compute eq = %modeleqn(mdl, 1)
compute post_coeffs_from_eq = %eqncoeffs(eq)
display post_coeffs_from_eq

I get

Code: Select all

1.14120      -0.02528      -0.14643       0.02425      -0.46959      -0.55113       0.29368       0.27940       0.00000

which is correct, so it looks like the problem lies in my usage of LINREG(create, ...)

Posted: **Tue Feb 02, 2016 8:57 pm**

What are you trying to do with LINREG(CREATE)? The idea behind LINREG(CREATE) is to allow you to compute the coefficients and covariance matrix in a non-standard fashion. If you're just trying to show the altered coefficients, just do

disp eq

The RATS Software Forum

LINREG with create option zeros out all coefficients

LINREG with create option zeros out all coefficients

Re: LINREG with create option zeros out all coefficients