Our research questions suggest a methodology for analyzing
marketing effects on sales and aggregate retailer performance
(sales revenues and profits), while accounting for potential
marketing endogeneity. Therefore, we chose the persistence
modeling approach (Dekimpe and Hanssens 1995), which
has previously been applied to long-term marketing effectiveness
for conventional products (e.g., Nijs et al. 2001;
Pauwels, Hanssens, and Siddarth 2002), offering a basis for
comparison. This approach involves four steps. First, unit
root and cointegration tests investigate whether the performance
and marketing variables are stationary, evolving, or
cointegrated (Enders 2004; Johansen, Mosconi, and Nielsen
2000). Second, using the test results, we estimate a vector
autoregressive (VAR) model or a vector error correction
(VEC) model (Dekimpe and Hanssens 1999). Third, we compute
impulse response functions, which track the effect of a
marketing variable on the performance variables of interest
over time (Pesaran and Shin 1998). Fourth, we perform a
weighted least squares regression of the estimated long-term
elasticities on the product category factors, using the inverse
of their standard errors as weights (Srinivasan et al. 2004).
The econometric specifications have been well documented