Fixed effects and SPSS

#1
Hi all,

For my bachelor thesis I have to calculate the sum of some betas (betas of x1, x2, and x3) of a linear regression model with fixed effects using SPSS 19 (STATA is only used in master degree). It is the following model:
y = x1 + x2 + ... + x6 + time dummies + fixed bank effects

y is revenue, where x1 - x3 are costs, and x4 - x6 are included to control.

But I cannot introduce fixed bank effects by creating new dummies, since this would take too much time (the data sets require at least 150 dummies, one for each bank name).
I figured out that fixed effects can be included using(Analyze-->Mixed models), but I wonder what I've to include under Subjects/Repeated, etc.
The only output I get is output without a full regression equation.
Can anyone help?

Thanks in advance!
 

noetsi

Fortran must die
#2
I have not seen mixed models used in linear regression before, why exactly do you need to add 150 seperate banks as dummies? That is almost certainly to many variables for your data set. What is your categorical variable?

It seems like you would want to group these banks for both theoretical and practical reasons (what does knowing what individual banks do tell you)?
 
#3
My categorical variable is bank.
I have to include this to capture bank-specific effects (λi). These effects are not captured by the variables x1 - x6.
I think this must be possible without creating dummies myself, but I don't know how since the Statistics courses I had only covered cross-sectional data.

Attached you find the formula on which I based my model (from Guevara and Maudos, 2011).
 
Last edited:

noetsi

Fortran must die
#4
Well (without knowing the research you are doing) I can think of two possibilities. First you can think of a way to group your banks in ways that make theoretical sense. Say large moderate and small banks were going to have different effects central to your model. In this case you would not need to have one level for each bank - you would just group the banks into one of these levels. Of course there has to be some theoretical basis for doing this in your model and you need to know where the banks would fit in a level and neither may be the case.

If you are looking for the random effects of banks (that is how banks vary effects your results) then you could treat your dependent variable as nested inside banks and calculate a bank random effect with a multilevel model. Or alternately used random effects ANOVA to do the same thing (I have not worked with that method just read about it).

Having said that the article you cite says "fixed effects are introduced in order to capture the influence of other bank specific unobservable factors that may effect its revenue." so I may not understand at all what you are trying to do. That sounds more like and error terms than the random one I was suggesting (or the grouping approach which simply replaces the dummies and leads to a fixed effect result).
 
#5
Thanks for your answer.
Your first solution is to capture the size effect of a bank.
But this is already captured by the function I use, since ln(size) is included.

I think fixed effects need to be introduced, and not random effects since also other
journals stress bank fixed effects. For example the attached one by Claessens and Laeven (2010). The fixed effect (ai) only changes for banks as subscript i indicates.
Further on they mention: "We estimate the model for each banking system using OLS
with fixed bank-specific effects.We also include time dummies for each banking
market". What do you think?
 

noetsi

Fortran must die
#6
Well the fixed effect of the bank would be the slope of that bank. But as you noted this would require you to create one dummy per bank - which seems to me to be unworkable. You can't have a model with 150 plus variables (well you can, but its an awful idea).

My group solution was not really aimed at size. What I meant was that you could group the banks into categories based on the key dimension you want to measure with size simply being an easy example of this. To prevent you from having to create so many dummies.

Further on they mention: "We estimate the model for each banking system using OLS
with fixed bank-specific effects.We also include time dummies for each banking
market". What do you think?
That I really would need to see the actual article cited to know exactly what they mean. It's not entirely clear what they mean by "fixed bank-specific effects" here. In OLS I assume you mean the slope of the variables as I noted above, since random effects are not generally raised in that context unlike say ANOVA or multileval analysis.

It's amusing that they say OLS when they mean linear regression- that greatly annoys some posters here :p

You probably need someone like Dason or Jake to help you with this - beyond my experience.
 
#7
That I really would need to see the actual article cited to know exactly what they mean. It's not entirely clear what they mean by "fixed bank-specific effects" here. In OLS I assume you mean the slope of the variables as I noted above, since random effects are not generally raised in that context unlike say ANOVA or multileval analysis.

It's amusing that they say OLS when they mean linear regression- that greatly annoys some posters here :p
Here you can find the full article: https://openknowledge.worldbank.org/bitstream/handle/10986/8906/wps3481.pdf?sequence=1
The fixed effects are mentioned two times. One time at page 9 and another time at page 18.

To me it seems like fixed bank-specific effects have the same effect as a dummy.
But in the article dummies are only mentioned explicitly with regard to the time effects.
And like you say creating that many dummies in SPSS is undo-able.
So the question remains: how can I be able to capture the ai of every bank?
 
#9
hi,
i know that this post is old, but I am curious if you could do the fixed effect on spss?
i have the same exact problem, where i need to capture bank-specific effect, but can't figure that out.
i would appreciate your help!