Re: Non-stationary sales data
- From: dave@xxxxxxxxxxx
- Date: Fri, 14 Sep 2007 10:20:51 -0700
On Sep 14, 11:37 am, "doher...@xxxxxxxxx" <doher...@xxxxxxxxx> wrote:
I have a dependent variable that is sales of product A in store 1. I
have store 2 which sells two products B, and C. B and C both impact
the sales of product A in store one. However, all sales are non-
stationary and trending over time. If i regress B and C on A
(dependent variable), B&C have extreme VIFs (~14). Now this tells me
that I can't trust the coefficients from the regression. I have take
the approach of modeling each product alone vs. A. How can I combine
the B and C in a manner that I may disaggregate them and then report
the impact of B&C as individual products on the sales of A?
Thanks.
Josh
Josh,,
When you have time series data regressions/correlations as you are
computing them have no value as the errors form the model/equation may
have
1. autocorrelation
2. a mean that is variant over time ...perhaps remedied by
Intervention Variables such as Pulses or Level Shifts or Locl Time
Trends or Seasonal Pulses like a Saturday Effect OR Holiday/Event
effects
3. non-constant variance
4. time varying parameters.
Ordinary Regression Models were developed for cross-sectional data.
You have longitudinal data. You should consult a statistician rhat can
help you identify and build TRansfer Function Models which seemlessly
integrate
Regression Structure , ARIMA structure and Interevntion Detection
schemes.
HTH
Dave Reilly
Automatic Forecasting Systems
http://www.autobox.com
.
- References:
- Non-stationary sales data
- From: doherjo1@xxxxxxxxx
- Non-stationary sales data
- Prev by Date: Non-stationary sales data
- Next by Date: Help Regarding R
- Previous by thread: Non-stationary sales data
- Next by thread: Help Regarding R
- Index(es):
Relevant Pages
|
