Anna Mikusheva refines time-series econometrics tools to improve forecasting | Hacker News Reader

Anna Mikusheva refines time-series econometrics tools to improve forecasting

newsoffice.mit.edu

70 pointsuser_23571114 comments

I did my PhD in econometrics. Reading this article, it looks like it was a general interest article profiling a professor who just got tenure. So if people are asking "how is this relevant to me", "how can I use this" or "should I learn more", the answers are most likely "it's not", "you can't", and "no". Not trying to be negative here, it's just that there are dozens of people getting tenure in econometrics every year at a similar level, all working on things that might seem to have some relevance to what HN readers are working on. From a HN readers perspective, there is nothing unique or special that I see here.

Hi! I'm currently (next year) getting my Master's in econometrics. I got my Bachelor Econometrics almost cum laude (7.95). My thesis was on using computational linguistics (Latent Dirichlet Allocation) to extract information from financial transcripts, which was cool. Do you feel like your PhD has been 'worth it'? I've been struggling with this question if I want to do a PhD. I have always loved academia and learning new things but I dont see really myself with a career in academia, maybe as a backup plan (how stupid that might sound).

Sorry for bothering you!

YMMV but it's less rewarding that what you would think (5th yr PhD Fin here)

I think the most important test is "do you care deeply about or are fascinated by the subject in question". If the answer is yes, then a PhD is not a bad idea (conditional on ok job prospects, inside or outside academia). A PhD will almost never be a better decision than a Masters degree from a financial perspective. But it can result in a different and perhaps more interesting career even in industry.

If you are considering economics, one thing you have to understand is that econometrics is very oriented towards economics, and if you are more interested in statistics per se, you would probably be better off in a stats program. There is very little machine learning in econometrics even today.

Thanks for taking the time to reply back, I really appreciate it.

Would you recommend working hard as hell and focus on getting my Master's cum laude, or get in some work experience on the side?

How long did your PhD take? I recently talked with an actuary PhD graduate and he did his in 3 yrs in The Netherlands, I also spoke to a graduate who did his at Cornell in 5 yrs.

The lecture notes for Mikusheva's graduate time series class are available through OCW: http://ocw.mit.edu/courses/economics/14-384-time-series-anal...

The math behind her research is a bit daunting...the course notes aren't trivial, but there's more intuition presented there than if you try to dive into her publications.

addendum: just to add ---- the course notes give a pretty good picture of the background knowledge necessary to be a time-series econometrician. If you're interested in the subject, check them out.

Time series econometrics on HN? Wow. I have a PhD in Econometrics too. I have met her at seminars and she is a wonderful lady with some good quality publications. But this is a fluff piece and she doesn't have any seminal contributions to the field. I wonder if PCB Phillips has ever made an appearance on HN.

Here is a link to her MIT page that contains here papers. I tried skimming one, but couldn't follow it; however, Based in the article I'm more interested in econometrics and might pursue that study further.

http://economics.mit.edu/faculty/amikushe/papers

Econometrics sparked my interest in statistics in general. The field has really spent a lot of time working on problems of "serial correlation," (basically dependencies in time-based measures). One of my former committee members wrote a great introductory text to econometrics, I highly recommend it: http://principlesofeconometrics.com/

^ user name checks out (unless you really do share the same initials as Ronald A Fisher)

Another great book that I read every year for refresher is Nearly Harmless Econometrics. I wish there were something like that more in tune for time series tho

edit: clicked on the link... turns out I have the same book (3rd ed)! So who are you? Hill, Griffiths or Lim?

Thanks for the recommendation. :) Hill was my committee member. One of my best teachers.

Her time series course notes are [relatively] more accessible: http://ocw.mit.edu/courses/economics/14-384-time-series-anal... A less technical but less up to date time series reference is Cochrane's free book: http://faculty.chicagobooth.edu/john.cochrane/research/paper...

"Econometrics" is just "statistics for economic applications." Time-series plays a big role (because macroeconomics and finance) but self-selection is really the dominant thread. People tend to make choices that they expect to benefit from, and it is often important to account for that in estimation. If that sounds cool to you, definitely check out econometrics. Somewhat shameless plug, but there's a list of the free resources I'm aware of at http://www.econometricslibrary.org/

I always thought that the primary difference between econometrics and statistics was the focus on causality/endogeneity. But I guess that in most cases, the endogeneity is caused by self-selection, e.g. selection into a treatment group, so both viewpoints are valid.

That's a good point and I probably agree with you. "Self-selection" was my quick attempt to translate "endogeneity" into english :)

Reply on news.ycombinator.com