Logit and Probit Regression

RoaStudent1984 · Postby **RoaStudent1984** » Wed Jul 18, 2012 8:05 am

Can someone please tell me how to run a probit and logit regression with Eviews?
I have absolutely no clue how the program works, and need these regressions to finish my thesis.
Both Regressions need to be run on 363(Date of Data) *10 (variables) data items.

For the Probit Regression I have:

- 6 independent control variables and 4 dummy variables.
- Dependent variables are valuation models used by brokers (0=Multiple, 1=DCF)

For the Logit Regression I have:

- 6 independent control variables and 4 dummy variables.
- Dependent variables are valuation models used by brokers (1=Multiple, 2=DCF, 3=SOP)

Thanks in advance!

EViews Glenn · Postby **EViews Glenn** » Wed Jul 18, 2012 11:10 am

Do you already have your data in EViews? Which parts of the manual have you looked at?

RoaStudent1984 · Postby **RoaStudent1984** » Fri Jul 20, 2012 4:49 am

Do you already have your data in EViews? Which parts of the manual have you looked at?

Yes my data is in Eviews and I found in the manual that I have to select the binary estimation method from the equation objects.
When I enter my equation specification the displayed information shows NA for St.Error, Z-statistic and Prob.

EViews Gareth · Postby **EViews Gareth** » Fri Jul 20, 2012 7:01 am

Do you get a warning message on the equation output?

RoaStudent1984 · Postby **RoaStudent1984** » Mon Jul 30, 2012 1:46 am

Do you get a warning message on the equation output?

There is indeed a warning message:

Singular covariance - coefficients are not unique
Covariance matrix computed using second derivatives

My equations input is as follows:
valuation_model c volatility size growth leverage mkt_bk r_d @expand(industrysector) @expand(industrysubgroup) @expand(cyclical) @expand(broker)

The regression output gives no warning when I leave the dummy variables out of the equation...

startz · Postby **startz** » Mon Jul 30, 2012 6:27 am

Dummy variable trap.

RoaStudent1984 · Postby **RoaStudent1984** » Mon Jul 30, 2012 7:31 am

Dummy variable trap.

If I leave the constant out of the equation I get the following error message:

Non-positive likelihood value for observation 1

startz · Postby **startz** » Mon Jul 30, 2012 7:53 am

See what happens if you only use one @expand statement at at time.

RoaStudent1984 · Postby **RoaStudent1984** » Mon Jul 30, 2012 9:01 am

See what happens if you only use one @expand statement at at time.

It works when I only use one @expand statement at at time.
Does this effect the outcome of the regression if I have to perform them seperately instead of at once?

startz · Postby **startz** » Mon Jul 30, 2012 9:04 am

The problem is that you are falling into the dummy variable trap with multiple categories of dummies. Basically, you need to drop one dummy from each category.

RoaStudent1984 · Postby **RoaStudent1984** » Tue Jul 31, 2012 8:42 am

Shouldn't I then just rewrite the equation so that every dummy is seperate?

So @recode(MS,1,0) @recode(CS,1,0) @recode(3510,1,0) @recode(3520,1,0) etc?

startz · Postby **startz** » Tue Jul 31, 2012 8:43 am

No. Look at the @drop options in @expand

RoaStudent1984 · Postby **RoaStudent1984** » Tue Jul 31, 2012 9:30 am

I tried the following:
@expand(industry, @dropfirst) but then I get the same error message as in the beginning.

startz · Postby **startz** » Tue Jul 31, 2012 9:32 am

Try googling "dummy variable trap"

RoaStudent1984 · Postby **RoaStudent1984** » Thu Aug 02, 2012 3:00 am

Thanks for all the help! The probit model is now completed.

Next step is that I have to run a multinomial logit regression.
The dependent variable can be either one of three choices where 1 is the reference point.
The dependent variable is followed by a set of independent variables (regressors?).

How do I have to create the equation so that I get values for both 2nd and 3d choice in one single output model?

EViews.com