Welcome Guest [Log In] [Register]
This forum is used with the NationStates web-game designed and run by Max Barry. While not officially affiliated, this serves as the regional forum for the regions: Middle East, African Continent, American Continent, Asian Continent, and European Continent.

You're currently viewing our forum as a guest. This means you are limited to certain areas of the board and can "read only".

In order to get the most out of these forums, please become a member and read this guide - http://z3.invisionfree.com/nationstates/index.php?showtopic=3060


If you're already a member please log in to your account to access all of our features:

Username:   Password:
Add Reply
Strange econometric result
Topic Started: Nov 16 2007, 06:09 PM (427 Views)
Paradise
Member Avatar
Resident bureaucrat

Am I the only one with skills in econometrics here? I'm getting odd results in a heteroskedasticity test and I don't know how to deal with it :unsure:

I'd need a second advice.
Offline Profile Quote Post Goto Top
 
Rhadamanthus
Member Avatar
Legitimist

Sorry Paradise, I have no knowledge about such things.
Offline Profile Quote Post Goto Top
 
Draxis
Member Avatar
Captain
I have no idea what you are talking about. So I will figure that simply, no, no I do not have skills in the region you require.
Offline Profile Quote Post Goto Top
 
Paradise
Member Avatar
Resident bureaucrat

I emailed my director about that. He should be able to give me the answer I seek.
Offline Profile Quote Post Goto Top
 
New Harumf
Member Avatar
Bloodthirsty Unicorn
Huh? I mean, seriously, Huh??
Offline Profile Quote Post Goto Top
 
Paradise
Member Avatar
Resident bureaucrat

?
Offline Profile Quote Post Goto Top
 
New Harumf
Member Avatar
Bloodthirsty Unicorn
I was referring to "heteroskedasticity".

Huh??
Offline Profile Quote Post Goto Top
 
Paradise
Member Avatar
Resident bureaucrat

One of Gauss-Markov's hypothesis was that the error term of a linear regression was to be homoscedastic, that is it its variance had to be a constant.

However, most of the time the error term variance is not a constant, it is a function of the independent variables. We then say that the error term is heteroskedastic.

Assuming the error term is homoscedastic while it isn't will make the inference incorrect. That is why a heteroskedasticity-robust standard error was created. However, typically this standard error is larger than the typical ordinary least squares standard error, which leads to less precise results.

Knowing if a model is homoskedastic or heteroskedastic is very important. So, I made this heteroskedasticity test for my HARCH (Heterogeneous AutoRegressive Conditional Heteroskedasticity) regression and I got very odd results. My Fischer statistic rejected null hypothesis of homoskedasticity, however, the p-values attributed to each coefficient were all insignificant. This is not supposed to happen. I should get at least one significant coefficient, otherwise, my F statistic would not reject the null hypothesis! That is VERY odd...
Offline Profile Quote Post Goto Top
 
Paradise
Member Avatar
Resident bureaucrat

Here's my test.

Prob > F = 0.0000 means the model is globally significative. In other words, it means I can explain the error term variance by my independent variables, which means I have heteroskedasticity.

However, if you look at the P>|t| column, you see that NONE of them are inferior to 0.100 which is the minimum to have significance on a 90% confidence interval.

Weird isn't it?

Code:
 
. reg sqres sqreturn1month sqreturn4monthnorm sqreturn5monthnorm sqreturn8monthnorm sqreturn10monthnorm sqreturn2yearnorm sqreturn3yearnorm sqreturn8yearnorm sqreturn10yearnorm in 313/972,
> noconst

     Source |       SS       df       MS              Number of obs =     660
-------------+------------------------------           F(  9,   651) =   12.41
      Model |  .001531162     9  .000170129           Prob > F      =  0.0000
   Residual |  .008921718   651  .000013705           R-squared     =  0.1465
-------------+------------------------------           Adj R-squared =  0.1347
      Total |   .01045288   660  .000015838           Root MSE      =   .0037

------------------------------------------------------------------------------
      sqres |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
sqreturn1m~h |   .0718839   .0634627     1.13   0.258    -.0527323    .1965002
sqreturn4m~m |   .0070831   .2558145     0.03   0.978    -.4952381    .5094043
sqreturn5m~m |   .0236732   .3195784     0.07   0.941    -.6038556     .651202
sqreturn8m~m |  -.2252533   .3836404    -0.59   0.557    -.9785753    .5280688
s~0monthnorm |   .4144089   .3707563     1.12   0.264    -.3136136    1.142431
sqreturn2y~m |   .0720352   .3709074     0.19   0.846     -.656284    .8003543
sqreturn3y~m |  -.1619877   .4171794    -0.39   0.698    -.9811672    .6571919
sqreturn8y~m |   .1729111    .768132     0.23   0.822    -1.335404    1.681226
sq~0yearnorm |   .6374083   .7341633     0.87   0.386    -.8042056    2.079022
------------------------------------------------------------------------------
Offline Profile Quote Post Goto Top
 
Tristan da Cunha
Member Avatar
Science and Industry
Statistics is not my forte. I would say watching TV is my forte.
Offline Profile Quote Post Goto Top
 
NRE
Member Avatar
Map Tsar and Southern Gentleman

My brain hurts now, thanks Paradise :P j/k
Offline Profile Quote Post Goto Top
 
New Harumf
Member Avatar
Bloodthirsty Unicorn
Paradise
Nov 16 2007, 11:32 PM
One of Gauss-Markov's hypothesis was that the error term of a linear regression was to be homoscedastic, that is it its variance had to be a constant.

However, most of the time the error term variance is not a constant, it is a function of the independent variables. We then say that the error term is heteroskedastic.

Assuming the error term is homoscedastic while it isn't will make the inference incorrect. That is why a heteroskedasticity-robust standard error was created. However, typically this standard error is larger than the typical ordinary least squares standard error, which leads to less precise results.

Knowing if a model is homoskedastic or heteroskedastic is very important. So, I made this heteroskedasticity test for my HARCH (Heterogeneous AutoRegressive Conditional Heteroskedasticity) regression and I got very odd results. My Fischer statistic rejected null hypothesis of homoskedasticity, however, the p-values attributed to each coefficient were all insignificant. This is not supposed to happen. I should get at least one significant coefficient, otherwise, my F statistic would not reject the null hypothesis! That is VERY odd...

Huh? You're making this up, right?
Offline Profile Quote Post Goto Top
 
Kasnyia
Member Avatar
Chairman of the Bank
Stop trying to make brain slushy, Paradise! :wacko: :blink:
Offline Profile Quote Post Goto Top
 
Paradise
Member Avatar
Resident bureaucrat

I don't know what to say. It is pretty simple for me :gnarkgnark:
Offline Profile Quote Post Goto Top
 
Kasnyia
Member Avatar
Chairman of the Bank
Yeah well, you stick to your bean countery and I'll stick to engineering and other mechanical stuff :D
Offline Profile Quote Post Goto Top
 
Nag Ehgoeg
Member Avatar
The Devil's Advocate

I know enough about economics to know that Carl Friedrich Gauss and Andrey Markov are real statitions and I can define homoscedasticity.

I don't know nearly enough to understand Paradises table. Like... at all. :blink:

So in short: nope, can't help you Paradise.
Offline Profile Quote Post Goto Top
 
Paradise
Member Avatar
Resident bureaucrat

Oh well. My director told me that multicollinearity between my regressors might be my problem. And well, he was right. Some of my independent variables are 94.35% correlated (this is VERY high) :o
Offline Profile Quote Post Goto Top
 
Catholic Europe
Member Avatar
Spammer
Am I tripping on acid or something? :wacko:
Offline Profile Quote Post Goto Top
 
New Harumf
Member Avatar
Bloodthirsty Unicorn
Catholic Europe
Nov 18 2007, 07:43 AM
Am I tripping on acid or something? :wacko:

If so, CE, I got some of the same stuff, and its a really bad trip! :wacko: :blink: :cry:
Offline Profile Quote Post Goto Top
 
Catholic Europe
Member Avatar
Spammer
Yeah, I don't like it.

Normally I get flying pink elephants.......not this time! :cry:
Offline Profile Quote Post Goto Top
 
Assassin
Member Avatar
Field Marshal
You prolly don't want to know what I got. :ph43r:
Offline Profile Quote Post Goto Top
 
flumes
Member Avatar
CLEVELAND ROCKS!
Someone remind me why I want to major in Business Econ???
:o :wacko: :unsure:
Offline Profile Quote Post Goto Top
 
Paradise
Member Avatar
Resident bureaucrat

flumes
Nov 20 2007, 05:23 PM
Someone remind me why I want to major in Business Econ???
:o :wacko: :unsure:

You're a geek and you love maths?
Offline Profile Quote Post Goto Top
 
Paradise
Member Avatar
Resident bureaucrat

I'm so happy. I will probably finish my Master's research report this week! It will make about 60 pages B)

By the end of this month, you will have to call me "Master" :P
Offline Profile Quote Post Goto Top
 
New Harumf
Member Avatar
Bloodthirsty Unicorn
Paradise
Dec 3 2007, 01:48 PM
I'm so happy. I will probably finish my Master's research report this week! It will make about 60 pages B)

By the end of this month, you will have to call me "Master" :P

Are you in a fraternity? Maybe the Betas?

If so, we can call you a "Master Beta"! :lol:

I'm already a "Master". Treat me with the respect I deserve! And my Master's Thesis was 110 pages long, with no formulas, spreadsheets or graphs to take up space! Plus, I wrote it on a typewriter, a manual typewriter, using white-out to correct mistakes - no word processors then!

Anyway, congrats and I look forward to you teaching soon!!

Also, didn't Andrei Markov do some work on optomizing baseball lineups for run productivity??
Offline Profile Quote Post Goto Top
 
ZetaBoards - Free Forum Hosting
Free Forums with no limits on posts or members.
Learn More · Sign-up for Free
Go to Next Page
« Previous Topic · Off-Topic · Next Topic »
Add Reply