Please use this identifier to cite or link to this item:
|Title:||Essays on merchant-generated contents and merchant-to-merchant spillovers in e-commerce platform||Authors:||Zhang, Xueli||Keywords:||Business::Marketing::Internet||Issue Date:||2021||Publisher:||Nanyang Technological University||Source:||Zhang, X. (2021). Essays on merchant-generated contents and merchant-to-merchant spillovers in e-commerce platform. Doctoral thesis, Nanyang Technological University, Singapore. https://hdl.handle.net/10356/147299||Abstract:||This thesis consists of three essays on merchant-generated content and merchant-to-merchant spillovers in the e-commerce platform. In Essay 1, I develop an analytic model to examine an online retailer’s optimal product quality display and pricing strategies in the presence of return options. Retailers have incentives to exaggerate their product quality to attract more consumers, but with the option of returns, retailers should reconsider it, since consumers might return the product if they find the real product quality is lower than the retailer-displayed quality. In the model, an online retailer offers a product with both vertical and horizontal attributes, sets a price, and displays quality to consumers. Consumers make the purchase and return decisions sequentially. I derive the retailer’s optimal strategies as a function of the real product quality, the consumer’s disutility factor for the product mismatch, and the disutility factor for the exaggeration. The main results show that the optimal price and displayed quality are negatively related to the exaggeration disutility factor but will not be affected by the mismatch disutility factor under certain conditions. Besides, the optimal profit will decrease as the mismatch disutility factor increases; and interestingly, it will first decrease and then increase as the exaggeration disutility factor increases. In Essay 2, I develop an AI system to extract features from both seller-posted and buyer-posted images and investigate the predictive power of the difference between these two types of images on product returns. Seller-posted images reflect consumers’ expected product quality, while buyer-posted images reflect the real product quality. A larger difference between these two types of images reflects a larger difference between the expected and real product quality, resulting in a high return rate. Specifically, first, I use well-established image-processing techniques to extract basic image attributes, such as entropy, cool color ratio, lightness, and saturation, which are defined as “shallow features”. Second, I use a residual learning framework, ResNet-50, combining with a special neural network structure, Siamese neural network, to recognize images, and choose the second-to-last layer of the network, containing 512 features, as “deep features”. Finally, I calculate the Euclidean and Cosine distances between seller-posted images and buyer-posted images using the extracted features, and I employ a state-of-art machine learning algorithm – gradient boosted regression tree – to predict product return rate leveraging the calculated distances. The results show that the estimated differences between the two types of images have statistically significant predictive power for product returns by increasing the prediction accuracy by 31.25%. In Essay 3, I develop an empirical framework to separately identify two opposite effects in e-marketplaces using consumer clickstream data. One is the positive spillover effect, which is captured by demand-side externalities, making firms tend to cluster together. The other one is the negative business-stealing effect because of the presence of competitors, hindering firms from co-locating in the same marketplace. I argue that the positive spillover effect is pointed identified, while the negative business-stealing effect is partially identified, or its bounds are identified. Then I non-parametrically estimate these two effects and explore their heterogeneities in different types of merchants. Besides, I examine how cashback rates will impact them separately. I find that cashback rates are positively related to the spillover effect and negatively related to the business-stealing effect. The effects of cashback rates are heterogenous in different merchant categories. Keywords: Merchant-generated content; user-generated content; analytical model; image processing; machine learning; deep learning; spillover effect; partial identification; clickstream data.||URI:||https://hdl.handle.net/10356/147299||DOI:||10.32657/10356/147299||Rights:||This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0).||Fulltext Permission:||open||Fulltext Availability:||With Fulltext|
|Appears in Collections:||NBS Theses|
Updated on Nov 26, 2021
Updated on Nov 26, 2021
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.