모델의 불확실성을 반영한 준거가격효과의 모형화

Abstract

The impact of reference prices on brand choice behavior has been modeled previously with both sticker shock and reference-dependent formulations. In the former model, reference prices impact choice via a "sticker shock" term, the difference between brand-specific reference prices and current prices while the latter model captures asymmetric responses to positive and negative Price deviations from a category-specific reference price. Though both models are well-grounded in psychological theory, they involve different choice and reference price formation processes, and impose different burdens on consumer memory. Both models been empirically validated in previous studies. However, previous empirical tests of these formulations have assumed that only one of these models holds for all consumers, ignoring the underlying structural uncertainty, i.e., the possibility that each process might be equally valid for different sets of consumers. In addition, an overwhelming majority of these studies have also either ignored price response heterogeneity, or modeled it using a latent segment approach, leaving open the possibility that unaccounted for heterogeneity may have biased the estimated reference price effects. This paper develops an integrated model that incorporates structural uncertainty and also models parameter heterogeneity at the household level using a Bayesian hierarchical framework. Specifically, the model does not assume that every consumer follows the same decision rule. Thus, some consumers follow the sticker shock model, because they have sufficient cognitive capacity and motivation to recall and use brand-specific shelf-prices, while others follow the reference-dependent structure and use a single, category-specific, reference price to decrease their cognitive burden. Response heterogeneity is modeled using a continuous random effects model that produces household-level coefficients and avoids the problem of residual segment-level heterogeneity. The model is estimated on scanner panel data from the ketchup category and outperforms hierarchical Bayes versions of the "pure" reference dependent model and the sticker shock model. On average, half the sample households in the ketchup category are found to be well described by the reference dependent structure and the rest by the sticker shock structure. Interestingly, the mean sticker shock coefficient is statistically indistinguishable from zero, i.e., there is no evidence for the existence of a sticker shock effect even after accounting for structural uncertainty. Overall, the proposed model provides a richer and more accurate description of the consumer choice process than simpler models that ignore model uncertainty and permit only a single model structure.