Quantcast

Question Time varying parameters (Master thesis Marketing Research)

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Question Time varying parameters (Master thesis Marketing Research)

jeroenvanvugt
This post has NOT been accepted by the mailing list yet.
Dear Statalist,

At the moment I got a wonderful assignment for my Master Thesis Marketing Research at a retailer. They want to know the effects of various promotions on sales in units (for various product lines). A return on investment analysis.

Excited as I was, I already started looking for the method I would use. What I noticed:

1. The sales is evolving for most of the products.

2. The effect of some promotions changes.

3. The promotions can have a lasting effect (hysteresis).

I have three questions:

1. It seems reasonable to use a regression model with time varying parameters for the promotions. In STATA I found two options: ARIMAX models or state space models. I prefer ARIMA because it’s easier to use. What syntax should I use and what literature should I read? I don't want to use the OLS regression:
Y(t)=constant+B1X1(t)+B2X2(t)+B3time+B4X1(t)*time+B5X2(t)*time…
because there are many promotions (more than two) and then there is multicollinearity.

Depending on your answer:

a. If you say state space models:
Can you indicate what the syntax would be for a state space model where:
y(t)=constant+B1(t)X1(t)+B2(t)X2(t) and B1(t)=B1(t-1) B2(t)=B2(t-1) where each X# is a different promotion

b. I you say ARIMA:
Can you help me with the adjustments to the syntax and finding good and easy to understand literature?

2. What is the best way to estimate the long run effect (hysteresis) and the direct effects? Should I use two separate models (one with positive and negative cumulative effects per promotion and one with direct effects) or can it all be estimated in one model?

3. Can I estimate ARIMA models for non stationary time series or should I use state space in which the time series is decomposed into various elements (trend etc.). What should I change in the model (what are the syntax adjustments).

If you can help me with pointing me towards a final overall answer then that would be great!

I have looked in various books and the stata manual but I'm still puzzled on how to implement this in Stata. Keep in mind that I’m not an econometrician.

Jeroen
Loading...