Archive for B-School Follies

Stathouse Letters

Model Search

I’m a simple, homoscedastic model and I’m not afraid to say it.  All I want is a nice primary dataset.  I’m not into multicollinearity or getting multiple independent variables together for wild regressions that iterate all night long and leave me nothing but a mess of residuals to clean-up.  Don’t get me wrong, I’m not looking for some tiny dataset with so much error it can’t even get a hat on its Y without extrapolating to back to other models about bikinis and swimming pools.  Just because I’m not into heteroscedasticity, doesn’t mean I don’t need a nice Z-score.  I’m just saying that a good data set isn’t all about size.  So if you have nice t-stats and aren’t ashamed of your p-value, come on over and give me a big F.  But don’t bother if your forecast doesn’t project well.  I need the total package; and for that dataset, I’ll have you smoothed and deseasoned before you can get one standard deviation away from the mean.  And then I’ll reseason you and do it all again.