Suppose I have this covariance matrix of regression coefficient for $\hat{\beta_1}$ and $\hat{\beta_2}$

\begin{pmatrix} &5 &-3\\ &-3& 0.5\\ \end{pmatrix}

And I'm going to test $\begin{cases} H_0:\beta_1+\beta_2=1\\ H_1:\beta_1+\beta_2\neq 1 \end{cases}$

If I am going to test with t-ratio I need to compute $se(\hat{\beta_1}-\hat{\beta_2}-1)$ so the square root of the variance:


This is an error depending on the covariance matrix which is not positive definite (in fact real part of the eigenvalues are not all positive). AM I right?

  • $\begingroup$ How did you get this covariance matrix? $\endgroup$ – EdM Oct 15 '15 at 16:21
  • $\begingroup$ On an exercise, but I think that there is obviously an error on the covariance matrix. $\endgroup$ – Marco Oct 16 '15 at 6:39

covariance matrix is always non-negative definite if your design matrix is full rank. Just compute your t-ratio as: t-value=(β1+β2-1)/sqrt([1 1] * covb * [1 1]')

| cite | improve this answer | |
  • $\begingroup$ 1) Every covariance matrix is positive semi-definite. 2) sqrt([1 1] * covb * [1 1]' is Negative! $\endgroup$ – Marco Oct 16 '15 at 6:38

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.