Re: Appropriate SAS Procedure to Analyze the Effect of Promotions on Sales to Business Customers
From: Chi Hing Ho (cccguy2000_at_yahoo.com)
Date: 08/19/04
- Next message: Kylie Veale: "Web-based Survey Sampling"
- Previous message: Scott Masten: "Power of Wilcoxon Matched-Pairs Signed Rank Test"
- In reply to: S. Kavanoor: "Appropriate SAS Procedure to Analyze the Effect of Promotions on Sales to Business Customers"
- Messages sorted by: [ date ] [ thread ]
Date: 19 Aug 2004 13:30:36 -0700
Hi,
This is a typical supply chain demand problem. The classical linear
model (including time series) and its extensions of standard
econometric models would not work in such senario. A good approach is
to use generalized linear models (GLM), with Poisson, Negative
Binomial or Gamma. Try PROC GENMOD in SAS STAT. You don't need to
worry much about the autocorrelation. However, for those intermitten
sales data, you need to model it as compound distribution, ie. find
the probability of no order, in this case a hidden markov issue.
C. Ho
kavanoor@aol.com (S. Kavanoor) wrote in message news:<66c55db6.0408040556.3f6e55a2@posting.google.com>...
> Hi All,
>
> Our client is a well known office supplies vendor and has about 50
> thousand business customers. Customer touch is through catalogs,
> direct mail, fax or field sales. Needless to say that the duration of
> impact varies depending on the type of touch. For example, catalogs
> which are valid for several months would have a longer term impact
> than faxes announcing Sale of the Week.
>
> We have customer level data for the last 30 months which includes
> sales (monthly) and the promotions they received and dates of the
> promotions. Sales frequencies vary considerably. While some purchase
> supplies every month others do not for several months.
>
> Our client wants to know how effective the different kinds of
> promotions have been and what the best promotional strategy is.
>
> We are planning on using SAS/ETS and are seriously considering impulse
> intervention model and distributed lag model of PROC ARIMA or PROC
> TSCSREG (Time Series Cross Section Regression). Can anyone advise us
> on the suitablity of these two (PROC ARIMA or PROC TSCSREG) techniques
> for the analysis of the data we have. If these two techniques are not
> appropriate what other technique(s), software should we look at?
>
> Any tips will be appreciated.
>
> S. Kavanoor
- Next message: Kylie Veale: "Web-based Survey Sampling"
- Previous message: Scott Masten: "Power of Wilcoxon Matched-Pairs Signed Rank Test"
- In reply to: S. Kavanoor: "Appropriate SAS Procedure to Analyze the Effect of Promotions on Sales to Business Customers"
- Messages sorted by: [ date ] [ thread ]
Relevant Pages
|
