Beyond mean estimates of price and promotional effects in scanner-panel sales–response regression

Traditional mean estimates of conditional sales given price and promotion variables may provide misleading guidance about the underlying market mechanisms, since high, low, and medium sales, respectively, may be generated by quite different price and promotion strategies. Empirical evidence for consumer good scanner data reveals nonlinearities and heteroskedasticity in the sales–response relationship—mean effects typically average and hence may obscure a potentially rich nature of observational data. Besides addressing the heterogeneity of price and promotional effects, the proposed quantile regression framework allows direct estimation of monotonicity restricted nonlinear pricing effects for quantiles of the sales distribution.