Leo,
When you model sales as it relates to advertising ( and other variables) it
is quite important to develop a model with any needed lag structures. One
has to make sure that and level shifts, seasonal pulses and/or local time
trends are incorporated/identified in addition to any one-time anomalies
which would distort parameter estimation and tests of significance.
Furthermore if there is residual auto correlative structure one needs to
remedy that to ensure that subsequent tests of significance are meaningful
statistically.
After constructing what might be an initial model one needs to validate
constancy of parameters over time and of course constancy of the variance of
the residuals over time before concluding the analysis.
There is a ton of material at
http://www.autobox.com/AFSUniversity/afsuFrameset.htm which might be of
help. In terms of textbooks I can highly recommend Wei, W. (1989). Time
Series Analysis Univariate and Multivariate Methods. Redwood City: Addison
Wesley. And http://papers.ssrn.com/sol3/cf_dev/AbsByAuth.cfm?per_id=328086
papers by Mike Hannsens . Furthermore
http://www.amazon.com/s?ie=UTF8&index=books&rank=-relevance%2C+availability%
2C-daterank&field-author-exact=Dominique+M.+Hanssens&page=1 would be a very
useful text for you to read.
After constructing models for each panel, the next step might be to test for
commonality of parameters thus leading to a possible segmentation pf panels
into homogenous sub-groups. This is sometimes referred to as pooled
cross-sectional time series analysis.
We at AFS have developed and are continuing to develop workhorse software to
help in these matters. One could always attempt to program these things in R
or S or even SPSS or SAS but as is often said "life is short but statistics
endures".
In terms of educational opportunities in this problem area , I am teaching a
3 hours seminar at the 27th annual ISF meeting.
http://forecasters.org/isf/workshops.html The ISF is charging $125 for this
course. AFS donates my time to the ISF in hope of improving not only the
theory but the practice of forecasting. Join me in NYC at the end of June.
In case that is not convenient, please feel free to call.
HTH
Dave Reilly
Automatic Forecasting Systems
http://www.autobox.com
215-675-0652
-----Original Message-----
From: A UK-based worldwide e-mail broadcast system mailing list
[mailto:[log in to unmask]] On Behalf Of Leo Guelman
Sent: Saturday, June 02, 2007 9:35 PM
To: [log in to unmask]
Subject: About the impact of Advertising on Sales
Hi,
I am trying to measure the impact of advertising (and other leading
indicators) on sales using panel data. If anyone worked on this problem,
could you share what is the right modelling approach to this problem? What
is the 'working-hourse' in the industry to this problem?
Many thanks in advance for your response.
Regards,
LG.
|