freecounterpoint: A spiffy Madoff man

Sunday, December 21, 2008

A spiffy Madoff man

11 comments:

Tecumseh said...: More Madoff stories: Markopolos waged a remarkable battle to uncover fraud at Madoff's operation, sounding the alarm back in 1999 and continuing with his warnings all through this decade. [He] reached his conclusion with the help of mathematicians like Dan diBartolomeo, whose analysis of the Madoff's methods in 1999 helped fuel Markopolos' suspicions. [..] Researching Madoff's numbers, using data the firm distributed to prospective investors, diBartolomeo concluded within hours that it was impossible for Madoff to get the returns he reported while using the strategy he said he used. "As the market goes up and down, this strategy should have done a little better or a little worse, just like everybody else," he said. "Instead, it appeared to be indifferent as to whether the market went up or down. They made money all the time."

You don't say!; 12/21/2008 9:34 PM
Tecumseh said...: Number crunching: "After spending about three hours playing around with regression analyses and various kinds of calculations, I could not reconcile it," diBartolomeo said. Hey, AA, what kind of regression analysis would have exposed the Madoff scheme?; 12/21/2008 9:39 PM
Arelcao Akleos said...: "More Madoff stories: Markopolos waged a remarkable battle to uncover fraud at Madoff's operation, sounding"

Ya Owe Me Tawny Port, AI [see "Only the Non-Ivy Rube Saw....", which just recently has scrolled off to "older post" land.]; 12/21/2008 10:53 PM
Arelcao Akleos said...: "Hey, AA, what kind of regression analysis would have exposed the Madoff scheme?"

Would have to know what the guy was doing regression on. I'd guess, based on the little he said, that he was taking performance data [return on investments] for a number of companies that were using the model Maddof SAID he was using and seeing if Madoff's data could be argued to be not a stark outlier. This would be a multivariate regression, where the cluster of data is related to an ellipsoid [usually just called the ellipse] containing most of it and for which any madoffian cluster lying far outside means "very unlikely to be so placed by randomness". This is a "linear regression", which is what almost all non academic types take to be regression [and most academic types to]. If it ain't linear looking at all, then they'll usually slap a log on the whole shebang to try and making it look more linear [hence "log-linear" regression]. Very few seem to have the willingness, or understanding, to handle properly problems where curvature says something important about the data.
Although there is a hell of a lot of grunt work in setting up multivariate regression problems, conceptually it ain't much other than similar to the movement from single variable to multivariable calculus. For stuff like this, separating the pile of dung from the wheat, the same ol same ol would have done just fine.; 12/21/2008 11:10 PM
Tecumseh said...: Oopsie daisy -- I missed that previous madoffian post, which I see went off the edge rather fast (who keeps drawing down the # of current posts? I bet it's JJ.) Note though that I added a link to the math type who actually did the calculation, Dan diB, who went to Cornell -- last I checked, Ithaca is an Ivy town.

I keep hearing about regression analysis, but never tried to do it myself on some real set of data. (My hunch is that >95% of the people who mumble the words wouldn't know how to do it.) What do you use for this -- SPSS?; 12/21/2008 11:33 PM
Tecumseh said...: PS: I changed the settings to have the latest 50 posts displayed, and selected a more detailed viewing scheme for the archive. Does this help? Hopefully, it will cut on those VCP/alz moments.; 12/21/2008 11:41 PM
Arelcao Akleos said...: SPSS is used mostly by the softer social science types [think education folk and not economists]. Probably the most common language to use is SAS, which has the same venerable place in statistical calculations as Fortran does for engineers. Academic types often prefer R, which is a tweaking of C+. There are specialized languages created by experts in the field, which are more convenient for doing deeper stuff, but rarely used outside of academia.
Actually, Matlab can do regression well. Mathematica? I dunno. Probably.; 12/22/2008 1:09 AM
Arelcao Akleos said...: ". Does this help? Hopefully, it will cut on those VCP/alz moments."

Yes, it helps. Thanks, AI. At the very least it should cut down on our accusing ourselves of getting alzheimers

[Hey, just read Columbo has alzheimers... a shame].; 12/22/2008 1:13 AM
Mr roT said...: Matlab is for homos.; 12/22/2008 3:03 AM
Arelcao Akleos said...: Freddy Mac is a Matlabl guy?; 12/22/2008 11:21 AM
Tecumseh said...: Mathematica is the real deal -- Matlab is for engineers. As for Pepe, he must use an abacus, non?; 12/22/2008 1:13 PM

Subscribe to: Post Comments (Atom)