Replies to This Discussion

Permalink Reply by Stefan Conrady on November 1, 2010 at 9:48pm

One of my favorite books in this regard is Applied Time Series Econometrics, http://amzn.com/0521547873 and JMulTi is a great software for multivariate time-series econometrics, which was created by the book's authors. It can be downloaded for free from www.jmulti.de.

Also, gretl is a great free time series tool (http://gretl.sourceforge.net/). Lee Adkins has written a free ebook specifically for time series analysis with gretl: http://www.learneconometrics.com/gretl/ebook.pdf. I hope this helps.

Stefan Conrady - www.conradyscience.com

• ▶ Reply

Permalink Reply by Yi-Chun Tsai on November 2, 2010 at 9:14am

Hi, Stefan:
Thanks. My background is in statistics & operations research and I want to know if I can teach myself some econometrics by reading either of the two books you mentioned above. Thanks.

• ▶ Reply

Permalink Reply by Yi-Chun Tsai on November 2, 2010 at 9:16am

Hi, Stefan:
Another question for you. I am trying to build marketing mix models and from your consulting experience, do you know any papers or webinar that covers this particular application? thanks.

• ▶ Reply

Permalink Reply by tom reilly on November 12, 2010 at 7:52am

Yi-Chun,

Those approaches listed above suffer from not addressing outliers like "pulse", "level shift", "local time trends", "seasonal pulses" in each of the time series being modeled. Without validating that the errors are NIID for each series these solutions are simply academic without industrial strength.

Check out Autobox(www.autobox.com) as we have a solution for this problem using our "Mixed Frequency Suite". See our Download page for more. We presented at the IBF in April using this approach, but for semi-hourly data. http://www.demand-planning.com/2010/03/18/can-forecasting-help-me-s...

• ▶ Reply

Permalink Reply by Tom Wolfer on November 12, 2010 at 4:46pm

Yi-Chun Tsai...what is 'vector' time series forecasting?....are you using upper and lower control limits? Is that what you mean? Thanks, Tom

• ▶ Reply

Permalink Reply by Yi-Chun Tsai on November 18, 2010 at 10:00am

Hi, Tom:
It is basically that you have several dependent variables such as y1, y2, y3, etc and you want to be able to model not only the dependency between those dependent variables but also betweem the dependent and independent variables. So, instead of building one univariate time series forecating model for each yi, where i=1,2,3,... , you want to do that simultaneouly. I am also just trying to learn this whole thing myself as we speak.

• ▶ Reply