Coefficient Sign Test

[diggetybo]

I know wald test typically deals with joint significance and magnitude, but...

Is it possible in eviews to perform a kind of wald test to test a hypothesis of whether a coefficient is positive or negative?

[trubador]

That's why null hypothesis tests against zero first. If you cannot reject it, then the sign does not matter since it is not significant at all. If you want to test against a specific sign, then you formulate a one-tailed alternative hypothesis. And the probability is equal to half of the value reported by EViews.

[startz]

Let me disagree with Trubador here (something that hardly ever happens...maybe we just disagree about what the poster meant be "or").

Testing if a coefficient is greater than zero calls for comparing the t-statistic to a one-tailed critical value (same for less than zero). For example, for a 5% test for a large number of degrees of freedom the critical value is 1.64 instead of 1.96.

If you actually want to test that a coefficient is either positive or negative, then that's the same as testing that it equals zero. In which case Trubador's answer makes sense.

[trubador]

Totally agree with Startz, and thanks for the clarification.

[diggetybo]

yea,

that does make much more sense now. thank you both trubador and startz.