# What is the difference between multivariate analysis and econometrics?

I am writing my thesis about how different factors influence stock excess returns after an M&A announcement. I haven't taken a multivariate analysis class and I have only a very basic idea of econometrics. I am not sure which one of those two will be what I need to learn to build my model.

In general, I will have one dependent variable (excess returns within specific period of time) and many explanatory variables, such as size of the acquirer, institutional ownership, different aspects of corporate governance etc. I want to see what influence, if any, each of those variables have on the dependent variable. What should I learn to be able to build my model?

To Expand: Econometrics includes Multivariate Analysis as a tool (a mathematical one). At the same time it may include many other things, such as economic "fundamental" models.

Econometrics is also a certain spin on (applied) statistics, just as biostatistics (one could say biometrics) or statistics in medicine, information theory or whatever field you can imagine. The unique problems faced in any field shape the tools needed for statistical analysis.

For example Econometrics is a very frequentist, non Bayesian field (at least how it is taught) - or at least one that doesn't teach two distinct approaches. The reason for that is that the data for most problems* lends itself to classic, frequentist analysis (lots of data points available, targets are constant mechanics). It is also a pretty standard approach in econometrics to learn OLS and then the General Method of Moments.
Panel Data models (Fixed/Random Effect) are, until needed, more of a fringe thing that comes up as a sidenote in most classes.
The data is generally considered continous. Logit, Probit and models which target nominal or ordinal variables are also not at the heart of what you learn in most Econometrics courses. Of course they are taught eventually and are important for economic fields which stray more into experiments or social areas but usually dependent variables are in money units.
Many medical people will learn things like ANOVA very througoutly. Econometricians consider this to be a spin on linear regression and maybe never learn it as a distinct thing.
Before those things come up, students will study time series analysis very througoutly, with Hamilton (1994) being the premier book in that area.

Even if you study econometrics to a graduate level, many methods on this site will seem foreign to you at first. It is a applied take on statistics that goes deep where it needs to and omits things that are not important in the field of economics.

In your case I would say that learning Econometrics is the right thing to do. It includes the stuff you need to analyze problems such as the one you posted.
My recommendation would include three books on three different levels.

The first one is Introduction to Econometrics by Stock and Watson. It is written for economics undergrads without further interest in statistics - just the basic stuff without going deep in the math. On the other hand it brings you - method wise - up to topics like cointegrated analysis.

Next up are Econometrics by either Hayashi or Green. These are the standard econometric graduate textbooks. If things are not proven or analyzed completely, they at least refer to the mathematical texts that do. Those give you a pretty good understanding of the usual econometric problem.

Last up is Time Series Analysis by Hamilton. It pretty much includes everything you could know at the point when it was published (1994), though it is more on a late graduate or phd level at least for (non quant-) economists.

• Very good overview! Although Bayesian approach is getting more and more interest in econometrics, especially in empirical macroeconomics and finance, though it is not usually taught yet in undergrad courses. May 3, 2013 at 10:41
• I come from a 100% frequentist background in econometrics and am currently trying to get a grip on the Bayesian approach myself. I read on this site that many consider it intuitive. I disagree. It seems nobody can really define what it is and what is doing differently. You need some grounding in the terminology of information science to even be able to read the wikipedia article. It goes so far that it is so unclear defined that I literally don't know if I am doing Bayesian stuff already (as I clearly learned some of it in basic stats). There should be a book for people like me. I feel so dumb
– IMA
May 3, 2013 at 11:25
• Example: I just read in a paper (Jaynes) that in one experiment the researcher was doing a hypothesis test over the means of two variables (ie. which one is better) with normal distribution. He made some mistakes but what Jaynes picks out is that he tests over the whole interval (including B>A, which is not supported by data) instead of restricting the hypothesis to just A=B and A>B beforehand (so that B>A is not a possibility).
– IMA
May 3, 2013 at 11:33
• This seems so amazingly arbitrary given the distribution of the variables is known. I mean there is a reason we pick this kind of interval which is we are unsure that the data reflects the truth. And given a known distribution, I don't understand how we can just cut out half of the distribution. I doubt Jaynes is in error but I don't get it. At the present, for me, you either come from a science that already does Bayes and are familiar with the terminology and procedure (mostly experimental sciences), or it is really unintuitive
– IMA
May 3, 2013 at 11:36
• I mean I can see the argument that we the hypothesis test leaves out some information ie. that the data would never support B>A, but given that the data itself is only one experiment or - given the distribution, we can not assume it is accurate, this is unsure information and calls to question the whole validity of his approach. Sorry for the derail ;)
– IMA
May 3, 2013 at 11:36

Econometrics is a specialized branch of applied statistics. Multivariate analysis is a branch of mathematics that has a lot of applications to statistics. For a great econometrics intro (at the beginning PhD level), I recommend Mostly Harmless Econometrics.

It doesn't cover everything by any means, but if you're starting from scratch with no stats background but with decent math skills, it'll get you started.