EViews.com

Hello,
I'm trying to set some basic notion from econometrics using Eviews.
I have been following some youtube lessons, and they got me all confused.
So first, my biggest uncertainty:

First Statement: For a good regression model one must have significant variables, which is indicated by p-value (last column on coefficients line). For one variable to be considered significant p-value must be less than 5% (0.05)
Second Statement: Always set H0, the desirable hypothesis and H1 the alternative (in fact what you don't wish for).
Third Statement: When you get p values of less than 5% one must say: We reject H0 (so in fact accept H1); When you get p values of higher than 5% one must say: We cannot reject H0, so we must accept it

Before actual question, please correct any of my first 3 statements.
Actual question: If i use a regression model (simple), and i get some coefficient with p value less than 5%, that would mean by 3rd statement to reject H0 and accept alternative. Any model 'desires' to have significant coefficients, so based on 2nd statement i will set H0: variable is significant and H1: variable is not significant. So my dilemma now is... it all contradicts itself ... i have to reject H0 under p value < 5% ... so in fact variable is not significant... but that contradicts 1st statement.
Which is it!?? Please help.

The hull hypothesis, H0, isn't the desirable hypothesis. Often, it's just that the variable has no effect and you're hoping to reject the null and show that the variable does have an effect.

I suspected an answer like that, so basically you're telling me not to set H0 = 'what i desire' but 'what i don't desire'.

That's where the confusion arise, because in the same lesson i watched, after we accepted the model to be good, one must verify that residuals are not auto-correlated. So the author set H0: Residuals not auto-correlated ('desired' after all), and H1: Residuals are auto-correlated). He than used Breusch-Godfrey Serial Correlation LM Test, and got a p-value for Chi-Square of = 0.0192 which is < 5% so reject H0 and say residuals are auto-correlated = not desirable.
Moreover, except p value of coefficients, all others tests he does (serial correlation, homoskedasticity, normal distribution) he sets H0 as desirable.
If i take your advise it means in the end i have no rule to set H0 ... or at least i'm very confused about what to set in H0 when i make tests

The way classical hypothesis testing works is that you ask whether there is evidence against the null hypothesis. If there is, then you choose the alternative. If there isn't, then either the null is true or the evidence isn't strong enough to distinguish. Technically what you "desire" isn't relevant. You might have a theory that says a variable has no effect, in which case your theory suggests that the coefficient should not be significant. Or you might have a theory that says a variable matters, which would suggest that the variable should be significant.

But in practice what you say is pretty much right.