SEE = SqRt MSE Think about driving down the highway, looking at a speed limit sign (MSE… “Mph Sign” which is “square”) If I don’t have my contacts on, the sign is big (blurry, imprecise) and I’m very uncertain about how fast I should go (dependant variable) If I do have my contacts on, I can SEE clearly, and the sign is small (clear, precise), I’m pretty certain about how fast I should go. Big SEE = Bad (uncertain abt dependant variable) Small SEE = Good (certain) Just something I thought of… might help you, might not, and I hope I’m not explaining something that’s blatantly obvious
Also… to remember how to get the t-stat for a slope coef (which is easy anyway), just remember “T-stats are a bunch of B/S” bullhsit t-stat is B (slope) over S (standard error)
Serial Correlation… I like how someone came up with Watson (from Sherlock Holmes) to detect Serial Correlation “serial killer” Also, to correct for serial correlation, you use the Hanson method Think about the band Hanson (mmm bop), on the cover of a kids cereal (serial) box
Conditional Heteroskedasticity… Heteroskedasticity sounds like a weird pagan ritual (Detect with Breusch-Pagan test) Now think about a heterosexual white guy you know (chances are, you are one)… so you use white-corrected errors for heteroskedasticity… keeping in mind this hetero white guy… Heteroskedasticity results in too many Type I errors. hetero white guys have a body part that looks like a I (use your imagination) But what the hell is a Type I/Type II error?? So the hetero white guy with the I is at a nightclub… he’s a really good guy, but he gets rejected by some hot chicks (type I error= rejecting a true (good) null (guy) Now there’s a really hot, super drunk chick at the same club… she’s so drunk she goes home with a toothless old pervert (TWO-thless)… she “accepted” a bad guy (type II error= accepting a false “bad” null)
The Hansen method not only corrects Serial Correlation, it also corrects Conditional Heteroskedasticity So not only is the band, Hansen (mmm bop), on the cover of a cereal (serial) box… but if you listen to them too much you’ll turn gay. Therefore, Hansen “corrects” your heteroskedasticity (makes it homo)
Conditional Heteroskedasticity correct with white corrected std errors Serial Correlation - Hansen but if you have both Serial Correlation, and Conditional Heteroskedasticity, then use Hansen
You have a creative mind Florida Gator…Thanks.
keep going…
Listening to Hansen corrects your heteroskedacity… Chances I forget what the Hansen is = 0% Thank you for that one.
Type I error - reject the null when it is true. Type II error - accept the null when it is false. Positive serial correlation means the errors are smallish (close together) --> denominator is small so t test is high, type I error Negative serial correlation means the errors are smallish (close together) --> denominator is high so t test is low, type II error One is more prevalent than the other. Do you guys remember which one?
Type I correct?
Making up weird little phrases and stories like this can really, really help you remember stuff. It might help if you come up with them on your own though. the more off the wall and/or vulgar the phrase, the easier it is to remember you can also do this to help remember names… when you’re introduced to someone, think of a celebrity with the same first name, and picture the person doing something the celebrity does… for example, a couple weeks ago I was introduced to a guy named brett… I pictured the guy with a bandanna on his head, rocking out like bret michaels.
QuantJock_MBA Wrote: ------------------------------------------------------- > Type I correct? I’m not sure, I’m asking
TheAliMan Wrote: ------------------------------------------------------- > Type I error - reject the null when it is true. > Type II error - accept the null when it is false. > > Positive serial correlation means the errors are > smallish (close together) --> denominator is small > so t test is high, type I error > > Negative serial correlation means the errors are > smallish (close together) --> denominator is high > so t test is low, type II error > > One is more prevalent than the other. Do you guys > remember which one? You probably meant to say errors are larger under -ve Serial Correlation> And +ve serial correlation is more prevalent.
Re: Standard Error of the Estimate new Posted by: Florida_Gator (IP Logged) [hide posts from this user] Date: May 19, 2009 05:24AM The Hansen method not only corrects Serial Correlation, it also corrects Conditional Heteroskedasticity So not only is the band, Hansen (mmm bop), on the cover of a cereal (serial) box… but if you listen to them too much you’ll turn gay. Therefore, Hansen “corrects” your heteroskedasticity (makes it homo) Dying at work reading this - hilarious if this comes up and anyone in London hears me singing…
Here is how I remember - Traditional,Money demand, Fed model, Gov model Remember : (T)he (M)other (F)u***cking (G)overnment -> (TMFG) first two are for equities/last two for FI Domestic Exposure: - + - + (think about the algebra - graphs - x axis - to +) For MNC here is the only one you need LTFAR (L)ocal Curr (T)emporal (F)unc curre (A)llcurrent ®eporting currency If F=R then scratch F A R and you have to use Temporal If F is not equal to R then Allcurrent method L, F and R diff then L->Temp->Functi->Allcurrent->Reporting For temporal only - all historical rates for I,F,INT,COGS,DEP Inventory Fixed assets Intangiblles COGS DEPreciation Next: Compre Income includes - CAP © - CTA and Cash Flow Hedge G/L (A)FS unrealized G/L - available for sale §ension expense