Zoom Logo

Public Lectures on Structural Econometrics - CEMMAP masterclass - Shared screen with speaker view
JAYANTI BEHERA
32:04
What does 'arg' indicate?
Arie Beresteanu
32:20
argument of
XIAOXIA SHI
32:31
It means "argument", arg min is the minimizer of the objective function.
Elena Pastorino
32:48
argmin is the value of the control that minimizes the function of interest
JAYANTI BEHERA
33:09
Thank you
Rubén Pérez Sanz
34:06
What if we consider other p´s? i.e. p=1 or p=3
XIAOXIA SHI
34:25
The professor is talking about this now:-) (p=1)
Rubén Pérez Sanz
34:44
listing...thanks
Arie Beresteanu
35:38
odd values (except 1) are problematic. You have to make sure that the Lp norm stays positive.
Wilfried Youmbi
36:57
I guess we can use any norms to minimize our SSR given that norms are equivalent in R^n right?
Cedomir Malgieri
37:40
What is the role of the h functions up to now?
XIAOXIA SHI
37:43
@Wilfried, actually they are not equivalent in this context.
XIAOXIA SHI
38:42
@Wilfried, note that the professor is talking about p=1 now. It's different from p=2 (which is OLS).
Elena Pastorino
38:45
@Wilfried: it depends on which features of the data you wish to capture... see what Bob is explaining now
Arie Beresteanu
38:54
The arg min using L1 norm is not equal to the ar min when you use L2 norm.
Wilfried Youmbi
41:45
Ok. Thank you all for the clarifications!
Elena Pastorino
41:59
@Cedomir: h general notation for the 1) notion of distance between the "dependent" variable and the model you posit to account for it and 2) the model you are positing
Zhuotong Xie
42:10
how should we pick tau in LAD estimation?
XIAOXIA SHI
42:52
@Zhuotong: LAD corresponds to quantile regression with tau=0.5
Zhuotong Xie
43:28
@Xiaoxia: thanks. is there any particular reason?
Cedomir Malgieri
43:33
Thank you Elena. Then, will them become more “relevant” for nonlinear estimators?
Elena Pastorino
44:03
@Cedomir: yes, exactly
XIAOXIA SHI
44:47
@Zhuotong, slides 10 explains. In general, you can run quantile regression with any tau you want. Typically, people do report many tau's at the same time.
XIAOXIA SHI
45:08
LAD is a special case of quantile regression with tau=0.5.
Zhuotong Xie
45:16
@Xiaoxia: I see. Thanks a lot!
Elena Pastorino
49:04
@Zhuotong: in practice it depends what patterns of the data you are interested in. Are you interested, say, in accounting for the median income in the US,? Or are you interested in "explaining" the income of relatively more disadvantage individuals, say, in the first quartile of the income distribution? These considerations guides your choice of tau
Elena Pastorino
49:26
guide... sorry for the typos
Zhuotong Xie
50:58
@Elena: This is very clear. Thank you!
Rubén Pérez Sanz
52:13
Let e(N)  (e1, . . . , eN )0 denote the vector of unobserved variables. In slide 14. is meant for each observation from 1 to N, isn't?
Arie Beresteanu
53:55
yes, e_i is the unobserved of observation I (and I=1,..,N)
Arie Beresteanu
54:21
i not I
Rubén Pérez Sanz
54:32
sure, thanks!!
Jung S. You
01:25:01
thank you!
Julio Galvez
01:25:06
Thanks!
Irvin Rojas
01:25:07
thanks!
Maria Alexandra Castellanos
01:25:07
thank you
Deniz Ozabaci
01:25:08
Thank you very much
Giuseppe Moscelli
01:25:08
thank you!
Sutianjie Zhou
01:25:08
thank you
Saheed
01:25:08
Thanks
Andrew Proctor
01:25:10
thanks!
Yuliya Kulikova
01:25:10
Thank you, Bob!
Vladimir Yankov
01:25:12
thank you!
Dante Langone
01:25:13
Before you mentioned t and s as different dimensions can you provide an example of that?
Andrea Flores
01:25:14
Thank you!
Wilfried Youmbi
01:25:14
Thank you!