Algorithmic Self-Preference in E-Commerce: Analyzing Amazon’s Buy Box Allocation Across Different Seller Types and its Antitrust Implications Public
Ni, Shuqian (Spring 2025)
Abstract
This thesis investigates algorithmic self-preferencing on Amazon by analyzing Buy Box allocation across different seller types. Key contributions include introducing referral fees and offer count as novel explanatory variables to capture Amazon’s commission-based incentives and platform competition intensity, and conducting a cross-country comparison with an underexplored market—Japan. Using product-level data from Keepa across the U.S., France, and Japan, we used logistic regression models with bootstrapped AME to evaluate how referral fees, pricing, market structure, and quality metrics influence Amazon’s probability of winning the Buy Box. Robustness is confirmed through LASSO regression and 4 alternative model specifications.
Findings show Amazon is more likely to win the Buy Box in high-referral-fee categories, suggesting a strategic incentive to prioritize market dominance over short-run commission revenue. Secondly, the negative association between the current Buy Box price and Ama- zon’s probability of winning Buy Box is weaker in high-referral-fee categories, revealing an internal trade-off between commission revenue and price margin. Thirdly, in low-referral-fee categories, Amazon is more likely to win the Buy Box without offering the lowest price over the past 90 days. This indicates stronger algorithmic favoritism in low-referral-fee categories, where the platform’s algorithm disproportionately favors its own retail offer in Buy Box allocation even when third-party sellers offer more competitive prices. Overall, this thesis offers new insights into the economic and strategic motivations behind algorithmic self-preferencing and underscores the need for more tailored, tier-sensitive antitrust regulatory responses.
Table of Contents
1 Introduction 1
1.1 Motivation..................................... 1
1.2 Buy Box and its Anti-trust Concerns ...................... 2
1.3 ResearchObjectives................................ 5
2 Literature Review 6
3 Hypothesis 10
3.1 Competition Intensity .............................. 10
3.2 Price........................................ 11
3.3 Quality Metrics-Reviews ............................ 12
3.4 Inventory Levels.................................. 13
3.5 90-days Sales Rank ................................ 13
3.6 Referral Fees ................................... 14
3.7 Geographic Differences .............................. 14
4 Data and Variable Description 16
4.1 Data Source and Collection ........................... 16
4.2 Data Cleaning................................... 16
4.3 Interaction Variables ............................... 18
4.4 Descriptive Statistics ............................... 19
5 Methodology 21
6 Results 23
6.1 Regression Model Setup ............................. 23
6.2 Main Variable Interpretations .......................... 23
6.3 Other Main Specifications ............................ 24
6.4 InteractionTerms................................. 26
7 Robustness 27
7.1 Self-preferencing Indicator ............................ 27
7.2 LASSO and Different Specifications....................... 29
8 Conclusion 31
8.1 Conclusion..................................... 31
8.2 Limitation..................................... 33
8.3 Antitrust Implications .............................. 34
Appendix 40
A.1 Data Cleaning Process .............................. 40
A.1.1 Raw Data Processing........................... 40
A.1.2 Binary Variable Cleaning......................... 41
A.1.3 Categorical Variable Cleaning ...................... 42
A.1.4 Scale Transformation........................... 43
A.2 Table1:Regression Variable Description .................... 45
A.3 Descriptive Stats ................................. 46
A.3.1 Mean Values of Main Variables Across Countries. . . . . . . . . . . . 46
A.3.2 United States Descriptive Statistics................... 46
A.3.3 France Descriptive Statistics....................... 47
A.3.4 Japan Descriptive Statistics ....................... 47
A.3.5 Distributions of Key Variables by Country . . . . . . . . . . . . . . . 48
A.4 Regression Results ................................ 51
A.4.1 United States Regression......................... 51
A.4.2 France Regression............................. 52
A.4.3 Japan Regression ............................. 53
A.5 Robustness .................................... 54
A.5.1 Self-Preferencing Frequency ....................... 54
A.5.2 LASSO Result............................... 54
About this Honors Thesis
| School | |
|---|---|
| Department | |
| Degree | |
| Submission | |
| Language |
|
| Research Field | |
| Mot-clé | |
| Committee Chair / Thesis Advisor | |
| Committee Members |
Primary PDF
| Thumbnail | Title | Date Uploaded | Actions |
|---|---|---|---|
|
|
Algorithmic Self-Preference in E-Commerce: Analyzing Amazon’s Buy Box Allocation Across Different Seller Types and its Antitrust Implications () | 2025-04-09 10:28:47 -0400 |
|
Supplemental Files
| Thumbnail | Title | Date Uploaded | Actions |
|---|