Nonstationary Brand Variables in Category Management: A Cointegration Perspective
指出品类管理模型因非平稳时间序列数据易产生伪回归问题，提出使用贝叶斯向量误差修正模型处理非平稳性，从而区分控制变量的永久与暂时效应，帮助决策者判断品牌变量对销售的短期或长期影响。
ABSTRACT Category‐management models serve to assist in the development of plans for pricing and promotions of individual brands. Techniques to solve the models can have problems of accuracy and interpretability because they are susceptible to spurious regression problems due to nonstationary time‐series data. Improperly stated nonstationary systems can reduce the accuracy of the forecasts and undermine the interpretation of the results. This is problematic because recent studies indicate that sales are often a nonstationary time‐series. Newly developed correction techniques can account for nonstationarity by incorporating error‐correction terms into the model when using a Bayesian Vector Error‐Correction Model. The benefit of using such a technique is that shocks to control variates can be separated into permanent and temporary effects and allow cointegration of series for analysis purposes. Analysis of a brand data set indicates that this is important even at the brand level. Thus, additional information is generated that allows a decision maker to examine controllable variables in terms of whether they influence sales over a short or long duration. Only products that are nonstationary in sales volume can be manipulated for long‐term profit gain, and promotions must be cointegrated with brand sales volume. The brand data set is used to explore the capabilities and interpretation of cointegration.