| Science Forum Index » Economy Forum » Econometrics and Machine Learning |
|
Page 1 of 1 |
|
| Author |
Message |
| Guest |
 Posted: Thu Dec 15, 2005 2:57 pm |
|
|
|
|
I find myself using linear regressions in all of my Econ papers. A
friend of mine suggested that I use machine learning algorithms like
MLP, Naive Bayes, KNN instead. What do you guys think? Is there a
reason that Econometricians use linear regressions so heavily for
modeling? My friend claims that using his algorithms will give me
better predictive accuracy. Is that true? Is that all I care about?
Thanks,
-Greg |
|
|
| Back to top |
|
|
|
| Dirk |
| Posted: Fri Dec 16, 2005 8:21 am |
|
|
|
Guest
|
I am not quite sure whether I understand you correctly but if you mean
learning algorithms in neural networks (NN), my opinion is this:
NN might fit your data much better, but that doesn't imply that the
predictive/forecasting properties are any better. Linear regression
models allow are straighforward and imply a structural form. nonlinear
regressions are even better and closely related to NN.
Dirk |
|
|
| Back to top |
|
|
|
| Phil Scott |
| Posted: Sun Dec 18, 2005 6:31 pm |
|
|
|
Guest
|
"Dirk" <dirknbr@googlemail.com> wrote in message
news:1134739292.308116.162950@g14g2000cwa.googlegroups.com...
[quote:1aa87c2427]I am not quite sure whether I understand you correctly but if
you mean
learning algorithms in neural networks (NN), my opinion is
this:
NN might fit your data much better, but that doesn't imply
that the
predictive/forecasting properties are any better. Linear
regression
models allow are straighforward and imply a structural form.
nonlinear
regressions are even better and closely related to NN.
Dirk
[/quote:1aa87c2427]
Game theory is relevant also... Sim City is good practice on
the economic/ demographic issues.. you can see how the cause
chain reacts to produce unpredictable effects.. this is the
point commonly missed by economists. imo
Phil Scott
> |
|
|
| Back to top |
|
|
|
|
