Extend economics thesis with multivariate analysis

[MaartenvK]

Dear all,

Working on my Master Thesis I got the remark from my supervisor that I have to extend the research with a multivariate analysis. I have been looking at literature, text books and different forums how to approach this, but could not find any usefull sugestions.
If anyone could help me with setting up a basic multivariate analysis I will be truely gratefull!

The subject of my thesis is in short an analysis of who stockprices of banks in the period 2007 - 2014 react to characteristics of this banks year end 2007. The independent variables used are for example the efficiency ratio, loans/deposits ratio, average risk weigt, capital ratio, log of assets, return on assets, leverage ratio, average interest income, non performing loans ratio.

I made simple linear regressions for all these seperate variables, but now I am struggling to combine them in a multiveriate regression.

Thanks for your time and help, it is highly appreciated! (also references to literature or other threads are welcome)

Kind regards Maarten

 

[desireeeve]

use xlstat addin in excel, they have 30 days free trial, I think it's a bit convenient to use than R. They have different multivariate regression analysis available which is easy to use and they also have tutorials online.

 

[GretaGarbo]

Maybe the supervisor simply suggested to use a multiple regression and not necessary a multivariate regression.

A multiple regression is when you have several explanatory variables ("independent variables", x-variables) like:

y = b0 + b1*x1 + b2*x2+error

Conditional on the x-variables, the only variation will be in the y-variable, thus it is a univariate model.

A multivariate model is when you have several dependent variables, with one or several explanatory variables. (Or no explanatory variable. Think of a bivariate normal distribution. Then the mean for each of the variables are constant.)

This may sound as "playing with words" but it matters if you search on the internet. Also it would be formally wrong in your master thesis to call a multiple regression for a multivariate regression.

The usual thing is to run the multiple regression for each of the dependent variable. A multivariate multiple regression is seldom needed.

- - -

Is it possible at all to estimate something stable over that time period? Before and after the credit infarction/economic collapse? Does it exist a theoretical model where the parameters ("beta-coefficients") are constant. That is the fundamental assumption in a regression model - that the parameters are unknown but constant.

Maybe there is a lot of multicolinearity among those variables, I don't know. But please don't use stepwise regression.

Desireeevi is mentioning in another thread, partial least squares (pls). Maybe that can be something to use as a prediction model. Otherwise, if you have an idea about the relations between variables, maybe a structural equation model (sem) can be useful. This could be in the sense economists use sem and the way psychometricians use sem (following Jöreskog). There are several regulars here who are enthusiasts for sem.

- - -
Maybe xlstat addin can be useful. But it does not have a very good reputation among statisticians (correctly or incorrectly believed). It is interesting to hear that it is easy to use. But what do you do when the trial period has expired? Suppose that you that you just want to change one little thing when you are almost done with the thesis?

 

[MaartenvK]

Thanks for your reply and suggestions.
Indeed I was also struggling with multivariate vs multiple regression, there is only one dependent variable in my analysis (the change in stock price) that I am looking at two periods (short-term recovery, and long-term recovery) would not make it logical to combine these in one model.
With regards to your second remark, how can I check whether the beta's are constant? Which test do I have to perform for that? a simple t-test?
I will look into the SE model to see if I can incorporate it in my analysis.

Many thanks again!