Help with Financial Research -- Statistics

September 19, 2011 1:52 AM

I'm currently performing a mock research comparative analysis for a class. I'm researching the financials of 'our' metrics versus our highest competitors. One of the questions I'm attempting to answer asks if the variances (or standard deviations) of 'our' profit margins are statistically similar to the variances in profit margins of our competitors over a five year period. I have 11 companies representing the profit margins of our competitors, and of course the reports on our profit margins for these five years. Now it's been a while since my last stats course... so I'm wondering, would I use an F or Levene's Test to answer this? Or am I completely off base... and should look to another form of analysis.
posted by Steve073190 to Education (4 answers total)

Sounds like an ANOVA or multiple-ANOVA is what you want here. That will examine directly whether the variances are statistically different between the companies.
posted by FrereKhan at 2:26 AM on September 19, 2011

...Levene's test would also work for you, I think. The matlab vartestn function can test all the 12 companies simultaneously.
posted by FrereKhan at 2:31 AM on September 19, 2011

Response by poster: Hmmm... well I tried the =Ftest function for my two arrays, (the first being the profit margins for my company, and the second being the other 8--had to cut three companies based on revenue sizes--and calculated a rather odd p-value of .934. This seems a bit high for my initial computations with standard deviations of 3.89 and 3.98 for the my company and our competitors, respectively... why would they differ so much if the F-test compares variances of a population to a sample variance directly? ...especially since, judging from the naked eye, they look nearly identical.

I'll perform a Lavene's test as well... see what comes out.

(I'm fairly certain ANOVAs are for independent sample group analysis. And in this case, since I'm only comparing a sample to a population, it would by construction be a t-test. Also, I'm relatively certain one of the parameters for a one-way ANOVA is equal variances...? Initially that's why I disregarded it. But please, by all means, correct me if I'm wrong.)

Thanks for the help so far.
posted by Steve073190 at 3:04 AM on September 19, 2011

How seriously do you take this? Financial stat people have a whole enterprise on analyzing and comparing these kinds of data. For example, these series are all highly correlated. That violates the distributional assumptions of the test, but might be fixable.
posted by a robot made out of meat at 7:37 AM on September 19, 2011