Where's Dr. Bob when you need him?

I need to do a Durbin-h test for autocorrelation on some data but have missed a bunch of classes lately and have no idea what i am doing!

FUCK BEANS!

<BLOCKQUOTE><font size=2 face="Verdana, Helvetica, sans-serif">quote:</font><table border="0" width="90%" bgcolor="#333333" cellspacing="1" cellpadding="0"><tr><td width="100%"><table border="0" width="100%" cellspacing="0" cellpadding="2" bgcolor="#FF9900"><tr><td width="100%" bgcolor="#DDDDDD"><font size=2 face="Verdana, Helvetica, sans-serif">Originally posted by Adept:
have no idea what i am doing!

</font></td></tr></table></td></tr></table></BLOCKQUOTE>

ummmmmm sincerely,

That sports guy from CBS

I would be willing to bet significant amounts of money that i know more about my first post than you do!

Who really has no idea!?

<BLOCKQUOTE><font size=2 face="Verdana, Helvetica, sans-serif">quote:</font><table border="0" width="90%" bgcolor="#333333" cellspacing="1" cellpadding="0"><tr><td width="100%"><table border="0" width="100%" cellspacing="0" cellpadding="2" bgcolor="#FF9900"><tr><td width="100%" bgcolor="#DDDDDD"><font size=2 face="Verdana, Helvetica, sans-serif">Originally posted by Adept:
I ...have no idea what i am doing!
</font></td></tr></table></td></tr></table></BLOCKQUOTE>


<BLOCKQUOTE><font size=2 face="Verdana, Helvetica, sans-serif">quote:</font><table border="0" width="90%" bgcolor="#333333" cellspacing="1" cellpadding="0"><tr><td width="100%"><table border="0" width="100%" cellspacing="0" cellpadding="2" bgcolor="#FF9900"><tr><td width="100%" bgcolor="#DDDDDD"><font size=2 face="Verdana, Helvetica, sans-serif">I would be willing to bet significant amounts of money that i know more about my first post than you do!
Who really has no idea!?
</font></td></tr></table></td></tr></table></BLOCKQUOTE>
I'm not touching that bet. You win, you know more about nothing than anyone else here.

I had written a whole spiel about the Durbin-Watson d statistic, before realizing you had said h, which is for detecting serial correlation in autoregressive models.

Bummer. BUT, the h statistic is ten times easier to understand. Your null hypothesis is that there is NO first order (positive or negative) serial correlation.

h = (1 - 1/2(d)*sqrt((n)/(1-n(var(alpha)))

So you need to calcualte d first anyway. The good thing is that h is distributed asymptotically normal so the hypothesis test is basically the same as a z-test (+- 1.96 are your critical values at the .05 level).

<BLOCKQUOTE><font size=2 face="Verdana, Helvetica, sans-serif">quote:</font><table border="0" width="90%" bgcolor="#333333" cellspacing="1" cellpadding="0"><tr><td width="100%"><table border="0" width="100%" cellspacing="0" cellpadding="2" bgcolor="#FF9900"><tr><td width="100%" bgcolor="#DDDDDD"><font size=2 face="Verdana, Helvetica, sans-serif">Originally posted by Dr Bob:
The good thing is that h is distributed asymptotically normal so the hypothesis test is basically the same as a z-test (+- 1.96 are your critical values at the .05 level).</font></td></tr></table></td></tr></table></BLOCKQUOTE>

BBBWWWAAAHHHAAA!!! RACK!

You rule man.

I used the stat software to get the Durbin-Watson d, and then went through the steps.

I think i figured it out:

p = 0.0569935
Var (b1) = 0.133787
h = 0.164247

Null: not first order autocorrelation
Alt: first order correlation

h (less than) 1.96
0.164247 (less than) 1.96

Thefore, do not reject the null.

The assignment is too look at the relationship of bankruptcies in a specific sector. That test was a test of wheather lagged bankruptcies (the ln of) had an autocorrelation effect on period t bankruptcies.

The second part was to creat a dummy variable regarding GDP to see if GDP had an effect on bankruptcies. That sounds like a simple regression, but i am sure it isn't. Should i use something like GDP growth rate?

i.e.

y = a + bx + Dx
bankruptcies (ln of) = a + ln GDP + GDP growth r?

Does that make any sense?

These things are only worth 2%, so i have som bullshit answer about 'intuition' and shit, but whatever!?

Thanks for the help!!!!!

* I had square brackets outside circle brackets but hostboad doesn't support it!?

Hostboard is being a real bitch!!! I had to throw mathematical convention out the window just to post this!!!

ρ = 1 – 0.5d
ρ = 1 – 0.5 (1.886013)
ρ = 1 – 0.9430065
ρ = 0.0569935

Var (b1) = [s.e.(b1)]2
Var (b1) = (0.365769)2
Var (b1) = 0.133787

h = ρ [ n / (1 – n Var (b1)) ]1/2
h = 0.0569935 [10 / (9) (0.133787)]1/2
h = 0.0569935 (10 / 1.204083)1/2
h = 0.0569935 (8.305078)1/2
h = 0.0569935 (2.881853)
h = 0.164247

Ho: not first order autocorrelation
H1: first order correlation

h (less than) 1.96
0.164247 (less than) 1.96

There - with mathematical convention. Damn arrow brackets!

That model makes sense (since you need to test both level and change of GDP).

Is there a good reason to do the log of bankruptcies or GDP (is the distribution that splayed)?

Blah, blah, blah. Take it to the college hoops forum.

Muriel! Where's my dinner! http://www.hostboard.com/forums/