高維度自我向量迴歸於零售業行銷與銷售績效分析

High-Dimensional VAR for Retail Marketing & Sales Performance Analysis

朱家輝(Ju Jia-Huei)

指導教授 ： 莊皓鈞 ； 周彥君

國立政治大學/商學院/資訊管理學系/碩士(2021年)

https://doi.org/10.6814/NCCU202101578

摘要

零售業者在規劃商品行銷策略時，多仰賴各種資料分析技術作為輔助，而近年來許多零售與行銷研究，常採用計量模型評估行銷效果、預測銷售表現，然而在零售業中，因商品數量多、行銷手法多元，資料呈現高維度時間序列樣態，其待估計參數量多，導致模型估計不易。
在本研究中採用向量自我迴歸模型建立估計系統，結合Lasso方法實證於真實零售業資料，並自估計結果解讀行銷效果，也發展各種零售業資料分析技術，提升實務應用價值。

此外，本研究中也設計蒙地卡羅模擬實驗，評估模型是否能確實反應行銷效果，並提出基於Cross-Entropy方法的隨機最佳化演算法，與基於Lasso方法的迭代估計演算法進行比較，探討在高維度零售資料樣態時兩種方法的特性，於零售與行銷計量研究方法論發展貢獻。

關鍵字

行銷 ； 向量自我迴歸 ； Lasso ； 高維度資料 ； 模擬 ； 最佳化

並列摘要

Many retailers rely on various data analysis techniques to help planning marketing actions.
In recent years, many retailing and marketing research often applied mathematical models to quantify the marketing effects and forecast the sales performances.
However, in the retailing industry, the large number of products with the diversified marketing mix strategies lead to the inherent high-dimensional issue. Specifically, the amount of model parameters would be extremely large, which makes it difficult to estimate.
In this study, we develop the estimation system based on Vector AutoRegression model combined with the Lasso method. Besides the explainable marketing results, we also aim to develop the additional tools which can provide more applicable information.

To assess whether the estimation procedure can recover the true marketing effects outside the analyzed samples, we conduct Monte-Carlo simulation experiments and proposed the stochastic optimization algorithm based on Cross-Entropy method. We further compare it to iterative Lasso estimation procedure and figure out the characteristics of two methods in the case of high dimensional estimation. Making contribution on the theoretical development of retailing and marketing research.

並列關鍵字

Marketing ； Vector Autoregression ； Lasso ； High-dimensional Data ； Simulation ； Optimization

參考文獻

Aaker, D. A. and Keller, K. L. (1990). Consumer evaluations of brand extensions.Journal of Marketing, 54(1):27–41.

Ailawadi, K. L., Harlam, B. A., C ́esar, J., and Trounce, D. (2006). Promotion profitability for a retailer: The role of promotion, brand, category, and store characteristics.Journal of Marketing Research, 43(4):518–535.

Benham, T., Duan, Q., Kroese, D. P., and Liquet, B. (2017). Ceoptim: Cross-entropy r package for optimization.Journal of Statistical Software, Articles, 76(8):1–29.

Bertsimas, D., King, A., and Mazumder, R. (2016). Best subset selection via a modern optimization lens.The Annals of Statistics, 44(2):813 – 852.

Bertsimas, D., Lamperski, J., and Pauphilet, J. (2019). Certifiably optimal sparse inverse co-variance estimation.Mathematical Programming, 184(1-2):491–530.