Does Amazon's Dual Role Weaken Marketplace Competition?

Abstract

Amazon's dual role, as both marketplace owner and first-party (1P) seller, gives it power over third-party (3P) sellers who sell similar items. This dual role can weaken 3P sellers’ ability to compete, possibly harming 3P sellers and consumers. We examine three aspects of marketplace competition. First, we examine price change dependencies. We find that 1P prices drop after either Buy Box (i.e., the “Add to Cart” or default sales box on Amazon's product page) prices increase or large 3P price increases occur; 3P prices decrease subsequently. Second, we analyze Buy Box seller selection because this is a critical conduit for demand. We find that both high 1P and 3P prices are penalized in Buy Box selection. Low-reputation and intermittent 3P sellers cannot win the Buy Box even at significantly lower prices. For some prices, the Buy Box favors 1P sellers over equal-priced 3P sellers and vice versa for other prices. Third, to see whether entry barriers weaken competition, we estimate a 3P seller entry model. Higher 1P prices are associated with more 3P sellers, suggesting low entry barriers. Combined, our results suggest that Amazon's dual role does not weaken competition in the marketplace. We discuss implications for marketplace participants, antitrust policy, and research.

Amazon Marketplace, launched in 2000, is now the dominant e-commerce marketplace, with about 37.6% share of online retail and about 10.4% of total U.S. retail sales. There are 180 million members (including multiple-customer households) in Prime, its fee-based membership program, and there are additional non-Prime consumers. Amazon has a dual role in its marketplace as both the owner and a seller (first party, or 1P) alongside third-party (3P) sellers. These 3P sellers sell similar or identical items to Amazon. The number of 3P sellers has been steadily rising, with 1.1 million active 3P sellers who now account for 62% of sales share (Statista 2025a).

Amazon's dual role can give it significant power, which it can use to weaken the ability of 3P sellers to compete with the 1P seller. The business press has accused Amazon of favoring itself—or self-preferencing—over 3P sellers in the default seller option, or Buy Box, that is, the “Add to Cart” box that appears prominently on the product page through which the majority of purchases (about 80%) occur (Angwin and Mattu 2016). The Federal Trade Commission (FTC) has sued Amazon for self-preferencing and alleges that Amazon puts pressure on 3P sellers to use its fulfillment services for which it charges them high fees, weakening their ability to compete with the 1P seller (FTC 2023). Researchers have built theory models showing how 3P sellers are weakened by Amazon's dual power (Anderson and Defolie 2024) and have found empirical support for Amazon self-preferencing in search results (Farronato, Fradkin, and MacKay 2023).

A contrasting view is that if Amazon weakens 3P sellers through its dual role, the economic logic and sustainability of a hybrid (of 1P and 3P) marketplace is weakened. Empirical evidence shows the share of 3P sellers growing over time. This evidence is not consistent with the weakening of 3P sellers. From Amazon's perspective, weakening 3P sellers will reduce the very profitable fees it collects from them. Additionally, Amazon will bear the entire inventory carrying costs incurred in 1P sales, which do not exist in 3P sales. Researchers have examined these profit trade-offs for Amazon through theory models and offer a more benign implication of the dual role (Etro 2023; Long and Amaldoss 2024; Wang and Qiu 2024; Zou and Zhou 2025). Empirically, Gutierrez Gallardo (2021) and Lee and Musolff (2025) show that Amazon's dual role is helpful to consumers by offering them the right mix of 1P and 3P sellers, including through Buy Box selection.

Given these contrasting views, we study whether Amazon uses its dual role to weaken competition. The competition in the Amazon Marketplace and with external rivals like Walmart is complex. Therefore, we follow an empirics-first approach (e.g., Golder et al. 2023), which includes drawing insights from observed patterns in three interrelated questions. First, do 1P prices weaken competition? In Study 1, we examine how Amazon's 1P price changes in response to 3P price changes and external competitors and how 3P sellers adjust prices subsequently. Second, does the Buy Box promote or inhibit competition? In Study 2, we examine the Buy Box's role in setting marketplace price and whether 3P sellers are competitively disadvantaged. Third, is the marketplace conducive to 3P seller entry? In Study 3, we examine whether entry occurs when 1P prices are high.

By answering these three questions, our research takes a more holistic view of competition compared with current literature. Additionally, in contrast to current literature, we include both internal marketplace factors and external competitors and the role of 3P seller heterogeneity in Buy Box winner selection. Therefore, we can reach more robust conclusions about the impact of Amazon's dual role on marketplace competition.

To answer our research questions, we use data from Amazon for the period September 13, 2017, through December 31, 2018, and from four medium-ticket consumer durables: electric cooker, deep fryer, microwave, and luggage. Our data include 3P seller features and price and product characteristics from 30 brands in these four categories. To incorporate external competitor price influence, we scrape prices from competitor websites like Walmart, Sears, and Home Depot on identical or very similar brands.

The key findings from our three-part analyses are as follows. First, we find that Amazon reduces 1P prices when the prior period's Buy Box prices are high (regardless of who won the Buy Box prior—3P or 1P seller). Primary 3P sellers (i.e., those with high reputation and steadier presence) sometimes increase prices substantially (approximately 10%–20% in our sample), possibly because external competitor prices increase. In response, 1P prices are cut in the short term, with subsequent lowering of 3P sellers’ prices (approximately 5% in our sample). These price changes suggest that Amazon uses 1P sales to promote price competition and lower prices. Second, our Buy Box analysis shows that the selection penalizes both high 1P and 3P prices. We find a mix of 1P and 3P sellers winning the Buy Box at any price. For some prices, the Buy Box prefers the 1P seller over equal-priced 3P sellers and vice versa for other prices, suggesting Amazon's preference for a hybrid marketplace. However, low-reputation and intermittent 3P sellers cannot win the Buy Box despite offering significantly lower prices. This restriction on these 3P sellers serves the preferences of the large base of Prime customers who pay fees for better service, including speed, reliable delivery, and returns. Third, we find that higher Amazon 1P prices are associated with increased 3P seller entry, suggesting that the marketplace does not create entry barriers to weaken competition.

While we have some evidence of self-preferencing (and the reverse evidence of 3P preferencing), our combined results suggest that Amazon's dual role does not weaken competition in the marketplace. Instead, it encourages lower 3P (and marketplace) prices, provides opportunities for reputable 3P sellers to compete via the Buy Box, and allows 3P entry when marketplace prices are high. Thus, Amazon has a preference for low prices (for both 1P and 3P sellers) and a hybrid marketplace. In sum, 3P sellers need not reflexively fear Amazon's power in this hybrid marketplace. This finding supports the growing theoretical literature about the benign effects of Amazon's dual role on marketplace competition and adds to the limited empirical literature that supports these theoretical claims.

Our findings have implications for marketing stakeholders and researchers. First, Amazon's dual role in its marketplace competition being associated with low prices and only promoting reputable 3P sellers suggests benign effects on consumer welfare. Second, our results also have mostly benign implications for 3P sellers, who can determine their chances of winning the Buy Box based on their prices and reputation and decide whether to enter the Amazon Marketplace. Third, these findings of mixed preferencing also reduce antitrust concerns compared with studies that find only evidence for self-preferencing. This finding also contrasts with current literature and suggests further research exploration. Finally, our results indicate that algorithmic collusion, discussed extensively in academic literature, may not apply to Amazon in its pursuit of lower prices in the marketplace. This presents another research opportunity.

We discuss whether our findings from our historical data might apply today. Algorithmic pricing, Walmart's growth, and greater antitrust scrutiny have altered the competitive landscape. We provide exploratory comments on the transportability of our results to this new landscape, including a replication of key results (i.e., Amazon's 1P price changes in response to Buy Box prices, external rival platforms, and own lagged changes) with newer data from September 2023 to January 2024. Our three-part analysis framework is portable to current data, and our results based on historical data can provide a benchmark and a foundation for analysis of today's Amazon Marketplace.

The remainder of this article is organized as follows. We first review related literature and then discuss the Amazon Marketplace setting. Next, we present our analysis and the results and insights. We discuss the implications for key stakeholders, robustness of the results, and transportability to current times. We conclude with a discussion of future research directions.

Related Literature

Our research is related to work on Amazon's dual role and pricing on the Amazon Marketplace. We review these strands of literature next.

Studies on Amazon's Dual Role

Theory papers in economics and marketing offer mixed views on how Amazon's dual role affects competition in the marketplace. Anderson and Defolie (2024) argue that Amazon can disadvantage 3P sellers by raising their fees and self-preferencing its 1P offers through search and the Buy Box (see also Raval 2023). Etro (2023) theorizes that Amazon prioritizes its 1P role only when it is more profitable than 3P sales. Recent literature in marketing offers similar intuition but explores new aspects of Amazon's dual role. Long and Amaldoss (2024) show that 3P sales can be more profitable for Amazon because of advertising revenue and higher fees. Wang and Qiu (2024) find that when a marketplace owner also acts as a 1P seller of either private label or brands, it tends to feature lower-priced, higher-quality products in the Buy Box. Zou and Zhou (2025) show that any self-preferencing by Amazon in search can spur more competitive 3P pricing. In other words, the harmful effects of Amazon's dual role on competition may be overstated.

Empirical research on Amazon's dual role has so far been in economics only. It also offers mixed or incomplete evidence. Farronato, Fradkin, and MacKay (2023) show that Amazon's private label products rank higher in search than similar-quality branded products sold by 3P sellers. Waldfogel (2024) finds similar results with European data that Amazon's private label listing advantage decreases after increased antitrust scrutiny. Chen and Tsai (2023) find evidence of self-preferencing in search for “frequently bought together with” branded products. However, none of these studies examine self-preferencing in purchasing or in the purchase-adjacent Buy Box. In contrast to these studies, two empirical papers find opposite results. Gutierrez Gallardo (2021) finds that 1P sales check 3P market power and that removing 1P sellers would hurt consumer welfare. Lee and Musolff (2025) show that Amazon chooses only to self-preference in offers that consumers prefer, improving consumer welfare.

Our work differs from these empirical papers in several ways. First, consider the scope of our 1P and 3P pricing analysis in Study 1. Unlike Gutierrez Gallardo (2021) and Lee and Musolff (2025), we examine the role of external competitors on price changes by Amazon and 3P sellers, addressing concerns of omitted variables. Our dynamic setting considers multiperiod price dependencies that static reduced-form analysis (Farronato, Fradkin, and MacKay 2023; Waldfogel 2024) overlooks and hence may misrepresent the nature of competition. Second, we examine self-preferencing in the Buy Box and for branded products. As Waldfogel (2024) notes, private labels are about 1% of search results in recent years. ¹ Therefore, the focus on self-preferencing of private labels in search in the literature above (Farronato, Fradkin, and MacKay 2023; Waldfogel 2024) is quite narrow. Our study of self-preferencing in branded products in the Buy Box is significantly more consequential to consumer purchase decisions than previously examined self-preferencing in display (or search). Importantly, we are the first to show evidence of both 1P and 3P preferencing (in the Buy Box); previous studies have only found self-preferencing (in search and display). We also examine the role of 3P seller heterogeneity in Buy Box winner selection. Finally, our three-part analysis provides a more comprehensive view of Amazon's dual role in marketplace competition.

Other Studies on Amazon Marketplace Pricing

Researchers have studied the effects of external competitors on the Amazon Marketplace. Aparicio, Metzman, and Rigobon (2024) show that Amazon competes with Walmart and offline stores, with local retail intensity affecting prices. Hunold, Laitenberger, and Thébaudin (2022) find that Amazon wins the Buy Box by undercutting competitors like Walmart. While these papers focus solely on effects of external competitors on Amazon's pricing, our work includes both external competition (e.g., multiperiod price changes by competitors like Walmart, Sears, and Home Depot) and internal marketplace factors (e.g., 3P sellers’ and Buy Box price changes). This allows for a more holistic view of marketplace competition. For example, we infer Amazon's low-price preference because of both external competitive pressure and the sensitivity of sales rank to within-marketplace competition.

Algorithmic pricing is another factor that might influence marketplace prices. Hansen, Misra, and Pai (2021) show that it can raise prices, whereas Musolff (2022) provides empirical support. However, Hanspach, Sapi, and Wieting (2024) find that algorithm adoption lowers Buy Box prices. Unlike these studies, we focus on Amazon's marketplace behavior, including but not limited to 1P pricing. Our results do not support collusion (see the “Model and Results” section for details).

Research Framework, Empirical Setting, and Data

Figure 1 provides an overview of our three-part research framework.

Figure 1.

Framework Overview.

The Amazon Marketplace

Amazon sells brands through both 1P and 3P sellers. In its 1P sales, Amazon controls price and inventory, fulfills orders itself, incurs all shipping and inventory costs, and retains all the profits from sales. Amazon also earns revenue from 3P sellers through referral, sales, and fulfillment fees.

Sellers may fulfill orders themselves (fulfillment by merchant) or pay to use Amazon's fulfillment services (fulfillment by Amazon, or FBA). Amazon offers a paid subscription service for customers—the Prime program—which provides faster delivery and other benefits and allows 3P sellers to serve Prime-eligible orders. Currently, Amazon also generates growing revenue from 3P seller and manufacturer advertising on the marketplace.

These fees collected from 3P sellers and lower inventory costs on 3P sales create incentives for Amazon to support these sellers. However, some of these 3P fees (like referral) are ad valorem, whereas others (like fulfillment) are per unit. We might expect that this mix of fees at various prices within a category will affect Amazon's preference for 1P versus 3P sales via the Buy Box and consequently how Amazon prices its 1P sales. In Study 2, we examine whether 3P sellers are disadvantaged and whether Amazon self-preferences as the 1P seller.

Consumers can see seller ratings, reviews, and product sales rank. In categories where Amazon participates as a 1P seller, a single default seller—featured in the Buy Box—appears on each product page. A proprietary algorithm determines the Buy Box winner, and roughly 80% of sales occur through this listing (Chen, Mislove, and Wilson 2016). Seller price and reputation are important selection factors (Gómez-Losada and Duch-Brown 2019). Consumers are likely to be price sensitive. In Study 1, we examine how Amazon might use its 1P role to keep prices low.

In addition to prices and inventory levels, 3P sellers can decide whether to delist from the Amazon Marketplace if they run out of inventory or if profits are low. They can relist when conditions are favorable. New 3P sellers can also decide to enter the marketplace if they see profit opportunities. In Study 3, we examine entry determinants, including Amazon 1P pricing, to see whether Amazon strengthens competition through 3P entry.

While Amazon is a retail leader, it has several online and offline competitors. These competitors are likely to affect how Amazon's dual role affects marketplace competition. For example, among these competitors is Walmart. Because Walmart is also an online marketplace, it provides 3P sellers with an alternative host, which may limit marketplace fees on Amazon. These external competitors may also increase the incentive to keep prices low in the Amazon Marketplace. Amazon may use its dual role to increase competition within its marketplace to lower prices. We include external competitors’ online prices as a proxy for these possible effects in the Amazon Marketplace.

Data and Descriptive Statistics

We examine four durable categories: electric cooker, deep fryer, microwave, and luggage (Table 1). To validate the generalization of our results, we also scrape data from a fifth category—board games—where demand is more transient, with new topical games frequently emerging (see Web Appendix A1). Our category choices are for the following reasons: (1) Amazon is a seller in these categories, so we can examine its dual role as marketplace owner and seller; (2) there are identifiable nonprivate label brands; and (3) prices are not too low (e.g., for many consumer packaged goods), and hence price changes are likely to be consequential for consumer decisions.

Table 1.

Marketplace Category Brands Studied.

Category	No. of Brands	Total No. of SKUsa	Brands
Electric cooker	14	20	Avalon Bay (AB), Breville (BR), Cosori (CR), Crock-Pot (CP), Cuisinart (CSN), Cusimax (CMAX), Elechomes (ELC), Elite Platinum (EP), Geek Chef (GC), Hamilton Beach (HB), Instant Pot (IP), Magic Mill (MM), Oyama (OYM), Pressure Pro (PPRO)
Deep fryer	6	7	CSN, EP, HB, Presto (PR), Secura (SEC), T-fal
Microwave	4	8	Frigidaire (FGD), Kenmore (KNM), LG, Panasonic (PNS)
Luggage	6	7	Delsey (DL), Olympia (OLY), Rockland (RL), Samsonite (SAM), Traveler’s Choice (TC), Reaction Kenneth Cole (RKC)

a

For most brands, we selected one SKU so that we have a SKU–brand one-to-one mapping. Some brands (e.g., IP in the electric cooker category) had more than one popular SKU with consistent price fluctuations during the observation period. For these brands, we scraped multiple relevant SKUs. See Table A1 for the brands with SKU numbers in parentheses.

We scrape hourly data from Amazon's website from September 13, 2017, through December 31, 2018. We also scrape price data from three competitor websites: Walmart, Home Depot, and Sears (see discussion in Web Appendix A1). These are leading e-commerce sites in our time period (Amazon and Walmart are in the top ten; see Ecommerce Guide 2024) and sites of different types of competitors (e.g., Walmart is bricks-and-clicks, Home Depot is a home improvement store). While much broader than the types of competitors studied in the current literature, this is not a definitive list of competitor sites and types. Therefore, this external competitor price variable should be seen as a proxy. We conduct a second scrape from September 24, 2023, through January 31, 2024, for both Amazon and the rival websites. During this second scrape, the Amazon Standard Identification Numbers and stockkeeping units (SKUs) that are no longer valid or are not sold are replaced by the nearest identical item in each brand. This second scrape is for validation purposes, so our main results are from the first scrape.

The frequency of within-day price changes is low (mean intraday changes for 3P sellers = .3775 and for Amazon = .30131). Therefore, we use the daily change as our unit of analysis. Appendix Table A1 summarizes brand characteristics. There is significant heterogeneity in brand ranks, prices, and sellers. From our data, 3P sellers vary in offer duration, ratings, reviews, distribution terms, and tenure on the marketplace. We model seller heterogeneity using K-means clustering (see Web Appendix B1 for details and Appendix Table A2 for seller classification). There are two main seller types across all categories. Established sellers with long sales history, FBA subscribers, and sellers with multiple brand offerings are present for more days in the marketplace. We term them primary sellers. Other sellers (e.g., small, new entrants) are on the marketplace intermittently; we term them secondary sellers. Across all categories, primary 3P sellers have higher ratings, lower deviations from Buy Box prices, and a higher number of days in the marketplace and make a higher number of price changes (see Figure 2).

Figure 2.

Primary and Secondary Seller Characteristics Across Categories.

This classification of 3P sellers on Amazon Marketplace is underexplored in the literature. As we find in our subsequent studies, the 3P seller classification is consequential for Amazon 1P price changes (Study 1) and Buy Box win likelihood (Study 2).

Models and Results

Study 1: Do 1P Prices Weaken Competition?

We study how Amazon and 3P sellers adjust their daily prices in response to each other and external competitors, among other variables.

Defining Price Changes

Our unit of analysis is the day, indicated by t. Assume that, for each category, there are K = {1, 2, …, K} seller types in the marketplace, with Amazon or 1P seller denoted by K. There is a total of B brands per category analyzed. Each seller type $\mathrm{k}\in \mathrm{K}$ may sell more than one brand b in each time period (out of B brands). At time t, p_bkt is the price of brand b charged by seller of type $\mathrm{k}\in \mathrm{K}$ . Therefore, the daily price change by seller of type $\mathrm{k}\in \mathrm{K}$ for brand b = 1, 2, …, B is

$${\mathrm{y}}_{\mathrm{bkt}}={\displaystyle \frac{{\mathrm{p}}_{\mathrm{bkt}}-{\mathrm{p}}_{\mathrm{bk},\mathrm{t}-1}}{{\mathrm{p}}_{\mathrm{bk},\mathrm{t}-1}}.}$$

(1)

Covariates Analyzed

Price changes by Amazon, 3P sellers, and external competitors

Amazon and 3P do not offer all brands, and not all sellers are present in the marketplace throughout the tracking period. We thus select brand-seller pairs in our dependent variable based on the days present in the marketplace and the frequency of price changes (see Web Appendix B2 for our brand-seller selection criteria). Figure 3 reports significant price change correlations with own past changes and those of internal rivals (i.e., Amazon or 3P sellers) and external competitors (e.g., Walmart), and Buy Box prices (for brevity, lags up to one period are reported here) across categories. Because price changes on external competitor sites are less frequent, we use the maximum daily price change across all brands within a category for each competitor site. Notably, lagged price changes by both within-marketplace players and external competitors are significantly associated with current-period price change responses of both Amazon and 3P sellers. Therefore, we include lagged daily price changes—up to three days prior—for 1P, 3P, and external competitors (Walmart, Home Depot, and Sears) in the price change regression model.

Figure 3.

Amazon, 3P, and External Competitor Price Change Correlations.

Product characteristics

Given the high fraction of sales through it, the Buy Box price is an important covariate. As shown in Figure 3, the price changes by Amazon and 3P made on a brand are correlated with its past period Buy Box prices. In addition, we include brand star ratings to capture perceived quality, sales rank and derived top and bottom brand indicators as a proxy for market share, and the number of sellers that have listed a brand to capture competitive intensity. For brands with multiple SKUs, we compute the mean, maximum, and minimum of each attribute across SKUs for a given period. For brands with a single SKU, we use the SKU-level values. All brand characteristics are included with a one-period lag. For each brand, we include number of product reviews and answered questions. We also include characteristics of all brands, including price-dormant ones.

3P seller characteristics

We include 3P seller type reputation and tenure metrics as well as 3P sellers’ order fulfillment services arrangement. To account for heterogeneity of sellers within a seller type, we include one-period-lagged values of the daily minimum, maximum, and mean characteristic across all sellers in that type. This also accounts for dormant sellers’ prices. Because any brand may be carried by more than one seller type, we include characteristics of other sellers (e.g., star ratings) who carry this brand.

Seasonality and other controls

We also include dummies for weekends, seasonal dummies for the holiday season, and Amazon-specific events such as Prime Day, Black Friday, and Cyber Monday. Appendix Table A3 summarizes all covariates.

Price Change Model

Because the previously discussed covariates are very large in number (>100), we use multivariate random forest (MVRF; Segal and Xiao 2011) to select a set of meaningful variables jointly associated with Amazon and 3P price changes. MVRF is suitable for jointly modeling multiple seller price changes and identifying associated factors, borrowing strength from more frequent outcomes to detect factors for less frequent ones. This flexible method can uncover nonlinear relationships between outcome and covariates without assuming a seller's objective function. See Web Appendices C1 and C2 for discussion of the algorithm and benchmarking against XGBoost and LASSO (least absolute shrinkage and selection operator).

The dependent variable of the MVRF is defined as the multivariate price change of

$${\mathrm{Y}}_{\mathrm{t}}=\{{\mathrm{y}}_{1\mathrm{t}},{\mathrm{y}}_{2\mathrm{t}},\dots .,{\mathrm{y}}_{\mathrm{Kt}}\},$$

(2)

where ${\mathrm{y}}_{\mathrm{kt}}=\{{\mathrm{y}}_{1\mathrm{kt}},{\mathrm{y}}_{2\mathrm{kt}},\dots ,{\mathrm{y}}_{\mathrm{Bkt}}\}$ is the price change vector for the k^th seller with the marketplace owner denoted by K, and the price change of brand b at time t given as y_bkt from Equation 1. Our MVRF algorithm identifies relevant brand and seller features and autoregressive lags (up to the third order). We deploy the selected variables in a time-series version of the generalized additive model (GAM; see Friedman, Hastie, and Tibshirani [2001] for an overview). ² This model is well-suited for our research question where parameterization of complex competition in the Amazon Marketplace may result in incorrect inference. Our price change equation is as follows:

$$\begin{array}{rl}{\mathrm{y}}_{\mathrm{bkt}}=& {\alpha }^{\mathrm{bk}}+\sum _{{\mathrm{k}}^{\mathrm{\prime }}=1}^{\mathrm{K}}\sum _{l=1}^{\mathrm{r}}{\beta }_{{\mathrm{k}}^{\mathrm{\prime }}l}^{\mathrm{bk}}{\mathrm{y}}_{\mathrm{b}{\mathrm{k}}^{\mathrm{\prime }},\mathrm{t}-l}\\ & +\sum _{\mathrm{k}=1}^{\mathrm{K}}\sum _{\mathrm{j}=1}^{\mathrm{J}}{\pi }_{{\mathrm{k}}^{\mathrm{\prime }}\mathrm{j}}^{\mathrm{bk}}\mathrm{Price}\mathrm{Change}{\mathrm{Dummy}}_{\mathrm{b}{\mathrm{k}}^{\mathrm{\prime }}\mathrm{j}}\\ & +\sum _{\mathrm{e}=1}^{\mathrm{E}}\sum _{l=0}^{\mathrm{r}}{\gamma }_{\mathrm{e}l}^{\mathrm{bk}}\mathrm{External}{\mathrm{Competitor}}_{\mathrm{e},\mathrm{t}-l}\\ & +\sum _{{\mathrm{b}}^{\mathrm{\prime }}=1}^{\mathrm{B}}\sum _{\mathrm{f}=1}^{\mathrm{F}}{\vartheta }_{{\mathrm{b}}^{\mathrm{\prime }}\mathrm{f}}^{\mathrm{bk}}\mathrm{Brand}{\mathrm{Feature}}_{{\mathrm{b}}^{\mathrm{\prime }}\mathrm{f}}\\ & +\sum _{{\mathrm{k}}^{\mathrm{\prime }}=1}^{\mathrm{K}-1}\sum _{\mathrm{m}=1}^{\mathrm{M}}{\phi }_{{\mathrm{k}}^{\mathrm{\prime }}\mathrm{m}}^{\mathrm{bk}}\mathrm{Seller}{\mathrm{Metric}}_{{\mathrm{k}}^{\mathrm{\prime }}\mathrm{m}}\\ & +\sum _{\mathrm{c}=1}^{\mathrm{C}}{\delta }_{\mathrm{c}}^{\mathrm{bk}}{\mathrm{Controls}}_{\mathrm{c}},\end{array}$$

(3)

where y_bkt is seller k's price change response on brand b at time t, and α^bk is the brand-seller-specific intercept for the corresponding GAM specification. To introduce the effect of past price changes, we include both autoregressive terms and competitor seller price change lags, y_bk′,t−l, where l = 1, 2, 3. Price Change Dummy_bk′j is a set of price change dummy variables (1/0) indicating when the one-day-lagged price changes by j = ±5%, ±10%, ±20% by seller k′ in brand b. We also include external competitor price changes (External Competitor_e,t−l), where l = 0, 1, 2, 3 denotes lag effects. BrandFeature_b′,f captures f = {1, 2, …, F} features of brand b′ in time t − 1, including those retained by our MVRF variable selection algorithm. Likewise, the relevant 3P-seller-specific metrics of the k′ type as of time t − 1 are captured by the variables SellerMetric_k′m, m = {1, 2, …, M}. We also add control variables for weekend and seasonality in the δ coefficients. During model training, we compared linear (parametric) and nonparametric forms and retained the linear form where nonparametric models did not improve fit. We also removed covariates that reduced model fit (i.e., R²).

Results

Figure 4 summarizes the key results across categories; tabulated results are presented in Web Appendix C4, Tables W9 and W12–W15. For instance, Figure 4, Panel A, reports the effect (coefficients and standard errors) of own-brand Buy Box prices on 1P and primary 3P price changes, shown as bar charts with the effect on the y-axis and the corresponding category brand on the x-axis; a negative (positive) coefficient is shown as a downward (an upward) bar. In all four categories, Amazon and primary 3P sellers drive the largest price changes (10%–20%), whereas secondary 3P sellers have limited influence.

Figure 4.

Key Results of Amazon and 3P Price Change Model.

We find that Amazon tends to cut 1P prices after prior period Buy Box price increases (see Figure 4, Panel A; this is true for all except Crock-Pot in the electric cooker category and Panasonic in the microwave category). It often lowers its 1P prices when primary 3P sellers raise theirs by approximately 10%–20% (e.g., Breville and Crock-Pot in the electric cooker category, T-fal in the deep fryer category; Panel D). Primary 3P sellers reduce prices subsequently (Panel D; true of all categories). Panel B shows that Amazon's 1P price cuts are also short-lived. Therefore, Amazon appears to use its 1P role to keep marketplace prices low by making short-term price cuts, after which 3P sellers also reduce their prices (see Study 2 for the implications of winning the Buy Box).

Table 2 reports the outcomes three days before and after 1P price cut episodes for brands with the largest price changes (approximately 10%–20%) by 1P and primary 3P sellers during our tracking period, to illustrate the shift in Amazon and 3P Buy Box win shares and prices, sales rank, and 3P behavior. The results show that following 1P price cuts, the Buy Box shifts in favor of Amazon on average, with lower prices and higher sales rank for the respective brand. In response, 3P sellers typically cut their own prices at an average magnitude of approximately 4.84% (across categories and brands from our sample).

Table 2.

Outcomes Before and After 1P Price Cuts.

		Amazon Buy Box Win %		3P Buy Box Win %		Sales Rank		Buy Box Price		3P Price Change %
Category	Brand	Pre	Post	Pre	Post	Pre	Post	Pre	Post	Pre	Post
Electric cooker	BR	65.19	74.90	34.81	25.10	12,150	11,353	$231.14	$226.36	6.96	−4.54
	CP	82.88	83.33	17.12	16.67	358	385	$27.40	$25.87	4.45	−6.80
Deep fryer	CSN	97.62	97.62	2.38	2.38	10,064	9,068	$81.83	$80.30	7.52	−6.09
	T-fal	90.97	90.97	9.03	9.03	1,032	959	$97.17	$94.28	.91	−.99
Microwave	PNS	90.00	93.33	10.00	6.67	6,573	3,589	$214.00	$202.37	.02	−.02
Luggage	OLY	94.44	100.00	5.56	.00	1	1	$27.11	$25.49	6.62	−8.67
	TC	100.00	100.00	.00	.00	8	6	$44.17	$40.72	4.73	−6.75

Notes: See Table 1 for brand abbreviations. “Pre” and “Post” refer to three days before and three days after 1P price cuts, respectively. The table shows the mean values across all such price cut episodes.

To measure interbrand associations, we include cross-brand price changes and product features (e.g., reviews, product star ratings) as covariates and find significant associations across categories and brands (e.g., see the regression for Breville in the electric cooker category in Web Appendix C4, Tables W10 and W11). However, we see mixed directions of interbrand 1P and 3P price changes, making it harder to ascertain the overall competitive effect. We conduct additional tests on sales rank stability (see Web Appendix C5, Tables W17–W20) and shifts in Buy Box win shares between Amazon and 3P sellers (see Web Appendix C6, Table W21) during weeks of frequent 1P price cuts. We also measure the correlation between Buy Box price dispersion across brands and frequency of 1P price cuts (see Web Appendix C7, Figure W9). Across these three analyses, we do not find conclusive evidence on the presence of interbrand competition or Amazon's product line pricing strategy.

In all four product categories, both Amazon and 3P sellers’ current-period price changes are significantly associated with multiperiod (two to three days) lagged price changes by external competitors such as Walmart, Home Depot, and Sears (Figure 4, Panel E). The correlation is positive for 3P prices and negative for 1P prices. ³ This positive correlation of 3P price changes with external competitors might explain their large price increases. Combined with the evidence from Study 2 that shows internal competition for the Buy Box, this finding indicates that external and internal rivalry coexist in the Amazon Marketplace.

In summary, we find associative evidence that Amazon uses its 1P pricing to promote low prices in the marketplace: It cuts 1P prices when the Buy Box price is high in the previous period and when primary 3P sellers substantially raise their prices. Amazon's 1P price cuts in response to large 3P price increases are short term, suggesting that its 1P role does not weaken price competition.

Study 2: Does the Buy Box Promote Competition in the Marketplace?

The Buy Box is a key conduit for default sales on the Amazon Marketplace. We saw in Study 1 that after a substantial price increase by primary 3P sellers, both 1P and 3P sellers decrease their prices, with a consequent decrease in Buy Box prices and improvement in sales rank (from Table 2). In this study, we examine how the Buy Box provides incentives for lower 1P and 3P prices and whether Amazon weakens 3P sellers through the Buy Box. We examine three aspects of the Buy Box algorithm. First, we examine whether 3P seller heterogeneity matters in Buy Box wins, that is, whether primary and secondary sellers have an equal likelihood of winning the Buy Box, conditional on price. Second, we estimate whether Amazon is favored in the Buy Box relative to the 3P seller, conditional on price and the consequent margin implications for Amazon. Third, we examine the relationship between Buy Box prices and product sales performance. This helps establish the need for lower Buy Box prices. Where relevant, we link our results to the findings from Study 1.

Study 2a: Does the Buy Box treat all 3P sellers equally?

As an illustration, we examine Breville in the electric cooker category. Amazon and both primary and secondary 3P sellers sell Breville consistently in our data, and thus the Buy Box is visible across a longer time panel. This larger number of Buy Box days enables us to robustly estimate the Buy Box winner selection mechanism (Studies 2a and 2b), including secondary seller implications. Table 3 shows that in our time period, the 1P seller wins the Buy Box on 64.7% of days, followed by a primary 3P seller at 22.7% and a secondary 3P seller at just .8%. As expected from seller clustering, secondary 3P sellers are intermittently present compared with primary 3P sellers.

Table 3.

Buy Box Win Frequencies of Amazon and 3P Sellers of Breville.

Seller Type	Buy Box Wins	Star Rating	No. of Days Seller Was Presenta	% Buy Box Winsb
Amazonc	121	5.0	187	64.7%
Primary 3P	42	5.0	185	22.7%
Secondary 3P	1	4.5	132	.76%
Other sellers	.85	4.8	38	2.21%

a

Data period: September 2017–December 2018.

b

Buy Box wins as a percentage of days that the seller made an offer on the SKU.

c

Amazon does not report its own star rating; we assume it to be 5 for purposes of the model.

We model Amazon's Buy Box policy using a random forest model based on features identified in prior research (Chen, Mislove, and Wilson 2016; Gómez-Losada and Duch-Brown 2019). These include Amazon and 3P price levels, their price difference to the lowest offer, seller ratings, FBA status, free shipping, and Prime eligibility (see Web Appendix D, Table W26, for summary statistics).

Study 2a: Results

Table 4 reports the predictive performance of our model of the Buy Box algorithm. The analysis gives a high recall of 98.6%, measured as $\frac{\mathrm{True}\mathrm{Positive}}{(\mathrm{True}\mathrm{Positive}+\mathrm{False}\mathrm{Negative})}$ , where the positive class denotes Buy Box losses. That is, our estimated policy can accurately predict 99.14% of actual losses. The precision of the model determined as $\frac{\mathrm{True}\mathrm{Positive}}{(\mathrm{True}\mathrm{Positive}+\mathrm{False}\mathrm{Positive})}$ is 96.4%. That is, of the total losses predicted by the model, 95.3% are true losses. The accuracy of our model, measured as the percentage of outcomes (win or loss) that the model can accurately predict, is 95.28%.

Table 4.

Confusion Matrix and Related Performance Metrics.

	Actual Losses	Actual Wins
Predicted losses	691	26
Predicted wins	10	35
Positive Class	Loss
Accuracy	95.28%
95% CI of accuracy	[93.52%, 96.67%]
Recall	98.6%
Specificity	57.4%
Precision	96.4%
Balanced accuracy	78.0%

With this robust Buy Box predictor, we evaluate whether all 3P sellers have an equal chance of winning the Buy Box, conditional on price. We simulate 3P prices undercutting Amazon 1P prices by $1 on days when both are present. When the primary 3P seller undercuts Amazon’s price by $1, our policy predicts that it wins the Buy Box on 98 out of 185 days (a 53% success rate). Therefore, the Buy Box selection mechanism strengthens the ability of well-priced and well-reputed 3P sellers to win. However, the secondary 3P seller (with lower ratings and fewer days on the market) fails to win the Buy Box on any day using the same rule and even when it makes deeper price cuts, for example, $20 below Amazon’s 1P price. In sum, the Buy Box win likelihood for 3P sellers depends not just on maintaining low prices but also on ratings and consistent presence in the Amazon Marketplace.

Study 2b: Does the Buy Box favor Amazon over 3P sellers?

We analyze whether there is any evidence of self-preferencing, that is, whether the Buy Box favors Amazon over primary 3P sellers (from Study 2a, we can defocus on secondary 3P sellers). We estimate a multinomial logistic regression model as a function of 1P and primary 3P prices as follows:

$$\begin{array}{rl}{\mathrm{Y}}_{\mathrm{b}{\mathrm{k}}^{\mathrm{\prime }}\mathrm{t}}*=& {\beta }_{\mathrm{b}0}+\sum _{{\mathrm{k}}^{\mathrm{\prime }}=1}^{\mathrm{K}}{\beta }_{\mathrm{b}1{\mathrm{k}}^{\mathrm{\prime }}}{\mathrm{Price}}_{\mathrm{b}{\mathrm{k}}^{\mathrm{\prime }}\mathrm{t}},\\ {\mathrm{BuyBoxWin}}_{\mathrm{bt}}=& \{\begin{array}{cc}1:& \mathrm{Amazon}\mathrm{wins}\mathrm{if}{\mathrm{Y}}_{\mathrm{bKt}}*=\underset{{\mathrm{k}}^{\mathrm{\prime }}}{\max }{\mathrm{Y}}_{\mathrm{b}{\mathrm{k}}^{\mathrm{\prime }}\mathrm{t}}*,\\ 2:& \mathrm{Primary}3\mathrm{p}\mathrm{wins}\mathrm{if}{\mathrm{Y}}_{\mathrm{bkt}}*=\underset{{\mathrm{k}}^{\mathrm{\prime }}}{max}{\mathrm{Y}}_{\mathrm{b}{\mathrm{k}}^{\mathrm{\prime }}\mathrm{t}}*,\\ 0:& \mathrm{Other}3\mathrm{p}\mathrm{wins},\mathrm{otherwise},\end{array}\end{array}$$

(4)

where Price_bk′t is the brand b price offered by seller k′ in time t (with K denoting the marketplace owner Amazon and k as the primary 3P seller). Here, brand b = Breville.

Study 2b: Results

Table 5 presents regression estimates. There are three key results. First, across a price range, both 1P and 3P sellers have a chance of winning the Buy Box when they price equally (see Figure 5). That is, there is no price point at which either the 1P or 3P seller wins the Buy Box with 100% certainty. This is strong evidence of Amazon's preference for having a hybrid marketplace. Second, for equal prices, there is evidence for both self-preferencing (below $250) and 3P preferencing (above $250). As mentioned in our “The Amazon Marketplace” section, the mix of ad valorem fees and per-unit fees, combined with 1P fees from manufacturers, may create a complex trade-off of profit margins on 1P and 3P sales for Amazon, conditional on equal 1P and 3P prices. Third, for unequal 1P and 3P prices, the negative own-price coefficients in Table 5 suggest that the Buy Box penalizes both 1P and primary 3P higher prices. Therefore, based on the results of Study 1, we can estimate the Buy Box win likelihood when primary 3P sellers substantially increase their price by approximately 20% and the 1P seller reduces its price by 5%. Without loss of generality, we pick a sample price combination after which 3P sellers increase their price. Initially, prices are $232 for 1P and $240 for 3P sellers. At these prices, Amazon's Buy Box win probability is 77%, whereas that of the primary 3P seller is 7%. When the primary 3P seller increases its prices by 20% (to $288) and Amazon responds by cutting prices by 5% (approximately $220), Amazon's Buy Box win probability increases to 99.8% and that of the primary 3P seller falls to 0%. That is, the 1P price cut after a primary 3P price increase effectively eliminates the primary 3P seller's chances of winning the Buy Box.

Figure 5.

Amazon and Primary 3P Win Probabilities at Equal Prices.

Table 5.

Regression Estimates of Amazon's Breville Buy Box Win Model.

	Amazon			Primary 3P
	Coef.	SE	p-Value	Coef.	SE	p-Value
Intercept	14.249	5.348	.008	−8.043	7.687	.295
Amazon price	−.125	.022	<.0001	.077	.029	.009
Primary 3P price	.068	.028	.015	−.044	.041	.280

An alternative scenario is if Amazon had raised 1P prices by 20% from its average $232 to approximately $278, while the primary 3P seller kept its price constant at its average price of $240. Here, Amazon's win probability would fall to .1% and that of the primary 3P seller would rise to 93.7%. That is, when primary 3P and Amazon prices are unequal, the Buy Box chooses the lower-priced seller. This shows why price increases by 3P sellers are short-lived.

In addition to this study, we explicitly link the Study 1 finding of episodes of 3P price increases and 1P price cuts with implications for Amazon's Buy Box wins. We model Amazon's Buy Box win share as a function of 3P price increases and 1P price cuts (see Appendix Table A4). The results corroborate the finding that 1P price cut episodes are positively associated with an increase in Amazon's Buy Box win shares (as seen in Table 2). These insights, combined with the Study 2a findings, suggest that the Buy Box rewards lower prices when offered by primary 3P sellers or by Amazon as the 1P seller. There is evidence for preferencing 1P over 3P sellers as well as the reverse, when both offer equal prices. ⁴

Study 2c: What happens when Buy Box prices increase?

Next, we investigate why lower Buy Box prices are important for Amazon's marketplace goal. In the absence of direct sales, we use the daily sales rank of each brand and the Ghose and Sundararajan (2006) equation to convert sales rank into sales (see Web Appendix D3 for details). Based on this, we regress the following sales model:

$${\mathrm{Sales}}_{\mathrm{bt}}={\beta }_0+\sum _{l=0}^1{\delta }_{\mathrm{b}l}{\mathrm{BuyBoxPrice}}_{\mathrm{b},\mathrm{t}-\mathrm{l}},$$

(5)

where Sales_bt represents the sales of brand b on day t, and BuyBoxPrice_bt is the Buy Box price of brand b on day t. We consider both the current and one-period-lagged Buy Box price of the brand.

Study 2c: Results

Table 6 reports the sales regression estimates across all key categories and brands in our analysis. Unsurprisingly, product sales are negatively associated with the Buy Box price across categories. ⁵ This helps understand why the Buy Box rewards the lower-priced seller regardless of whether the seller is 1P or 3P (Study 2b).

Table 6.

Daily Sales Regression Results.

	Electric Cooker
	BR			CP
	Coef.	SE	p-Value	Coef.	SE	p-Value
Intercept	146.826	23.796	<.0001	6,605.876	221.476	<.0001
Buy Box price	−.212	.114	.065	−209.850	11.326	<.0001
Buy Box price last period	−.148	.065	.024	4.569	8.310	.583

	Deep Fryer
	CSN			T-fal
	Coef.	SE	p-Value	Coef.	SE	p-Value
Intercept	189.166	22.545	<.0001	789.077	133.452	<.0001
Buy Box price	−1.144	.637	.075	−.059	1.681	.972
Buy Box price last period	−.365	.635	.567	−4.610	.986	<.0001

	Microwave
	KNM			LG
	Coef.	SE	p-Value	Coef.	SE	p-Value
Intercept	36.475	7.718	<.0001	414.993	42.261	<.0001
Buy Box price	−.117	.082	.153	−2.189	.290	<.0001
Buy Box price last period	−.003	.067	.960	.673	.141	<.0001

	Luggage
	OLY			TC
	Coef.	SE	p-Value	Coef.	SE	p-Value
Intercept	22,375.000	8,816.780	.0131	1,326.634	4,926.176	.788
Buy Box price	−287.479	385.679	.4582	82.667	152.653	.590
Buy Box price last period	−211.557	396.193	.595	−25.215	162.744	.877

Notes: See Table 1 for brand abbreviations. For brevity, we provide summary results for pairs of brands per category. For the luggage category, sales are estimated from subcategory sales rank, as category rank was not observed during scraping.

These results also help explain the finding in Study 1 of a 1P price cut following a substantial 3P price increase. Using the same Breville example, if a 3P seller increases the price from $240 to $288, and if it were to win the Buy Box, Breville’s daily sales rank would worsen from 213,468 to 245,544, and daily sales would fall by about 14 units (from the sales rank to sales conversion equation). Expectedly, this reduces Amazon's fees from 3P sales and makes it less competitive relative to external rivals. This explains Amazon's need to cut its 1P prices to keep Buy Box offers competitive.

To summarize, we find that seller quality (i.e., days on the market, reputation) is necessary, but not sufficient, to win the Buy Box. Low prices are also necessary, but not sufficient, to win the Buy Box. Expectedly, low prices help with higher sales, indicating Amazon's incentives to strengthen competition among sellers within the marketplace.

Study 3: Is the Marketplace Conducive to 3P Seller Entry?

Entry model

In addition to using its own 1P pricing and Buy Box, Amazon can use 3P entry to affect competition in the marketplace. If it wants to strengthen competition, it should permit entry when prices are high, including its own 1P prices. Because we do not observe Amazon's entry fees and other restrictions in our data, we estimate that the total number of sellers selling an item in each period is correlated with price changes by Amazon and primary 3P sellers. Seller entry could also be due to both price- and brand-related factors. As in Study 1, we use past period (i.e., one-day-lagged) price changes by Amazon and 3P sellers, own-brand features (e.g., star rating, sales rank), seasonality, and weekend controls as the covariates. We also include one-day-lagged prices of Amazon, Buy Box, and mean non–Buy Box 3P sellers. The GAM equation is given as

$$\begin{array}{rl}{\mathrm{n}}_{\mathrm{bt}}=& {\delta }_{0\mathrm{b}}+{\delta }_{1\mathrm{bK}}{\mathrm{Price}}_{\mathrm{bK},\mathrm{t}-1}+{\delta }_{2\mathrm{b}}{\mathrm{BuyBoxPrice}}_{\mathrm{b},\mathrm{t}-1}\\ & +{\delta }_{3\mathrm{b}}{\mathrm{NonBuyBoxPrice}}_{\mathrm{b},\mathrm{t}-1}\\ & +\sum _{\mathrm{k}=1}^{\mathrm{K}}\sum _{\mathrm{j}=1}^{\mathrm{J}}{\delta }_{4\mathrm{b}{\mathrm{k}}^{\mathrm{\prime }}\mathrm{j}}\mathrm{Price}\mathrm{Change}{\mathrm{Dummy}}_{\mathrm{b}{\mathrm{k}}^{\mathrm{\prime }}\mathrm{j}}\\ & +\sum _{\mathrm{f}=1}^{\mathrm{F}}{\delta }_{5\mathrm{bf}}\mathrm{Brand}{\mathrm{Feature}}_{\mathrm{bf}}+\sum _{\mathrm{c}=1}^{\mathrm{C}}{\delta }_{6\mathrm{c}}{\mathrm{Controls}}_{\mathrm{c}},\end{array}$$

(6)

where n_bt is the number of 3P sellers selling brand b on day t. Price_bK,t−1 is the one-day-lagged price charged by marketplace owner Amazon. BuyBoxPrice_b,t−1 and NonBuyBoxPrice_b,t−1 are the one-day-lagged Buy Box and mean non–Buy Box 3P prices, respectively. The remaining variables are similarly defined as in Study 1.

Results

Figure 6 summarizes key factors correlated with seller entry across categories. We focus on the coefficients (and standard errors) for Amazon's 1P prices and dummies for price increases (or decreases) by 5%, 10%, and 20% (see Web Appendix E, Tables W30–W34, for detailed results). Across categories, higher 1P prices in a prior period are associated with an increase in 3P sellers in following periods (Figure 6, Panel A). Likewise, when Amazon raises (lowers) 1P prices in the prior period, there are more (fewer) sellers in the next period (Figures 6, Panels B and C). We see in our data that Amazon raises 1P prices when 3P sellers delist and find evidence of this across all the categories we studied. This likely encourages 3P sellers to enter (or relist), thus reducing prices. From the Study 2 Buy Box analyses, lower 1P and 3P prices result in lower Buy Box prices and hence higher sales. Therefore, Amazon has the incentive to attract entry and strengthen competition in the marketplace.

Figure 6.

Key Insights from Seller Entry Model.

Key Insights from the Three-Study Analyses

Table 7 summarizes our key insights. The results suggest Amazon uses its dual role to strengthen price competition. Very high 3P prices do not go unchecked (Study 1). Subsequent 1P price cuts lead to reallocation of the Buy Box, which does not improve Amazon's margins overall (i.e., 3P margin loss is higher). In response, 3P sellers drop their prices; that is, the price increases are short-lived. Higher 1P prices also do not go unchecked (Studies 2 and 3). Therefore, first, we infer that Amazon has a preference for low prices from 1P and 3P sellers. This preference drives price competition and provides opportunities for 3P sellers who can keep prices low.

Figure 7.

Key Summary of Validation Data Results (September 2023 to January 2024).

Table 7.

Key Insights at a Glance.

Study 1 Insights
• Large 3P price increases (approximately 10%) and higher Buy Box prices are short-lived. • Amazon (1P) cuts prices after major 3P increases in Buy Box prices. • 3P prices typically fall approximately 5% following 1P price cuts. • Amazon’s Buy Box share increases by about 2.7% after 1P price cuts.

Study 2 Insights

• The Buy Box prefers the low-priced seller, 1P or reputable 3P. At equal prices, there is mixed preferencing between 1P and reputable 3P sellers. • Secondary 3P sellers have low chances of winning the Buy Box even at low prices. • Based on industry margin estimates, Amazon earns lower gross profits when it cuts prices to win the Buy Box over higher-priced 3P sellers. • An increase in Buy Box price leads to lower sales and worse sales rank.

Study 3 Insight
• 3P entry occurs when marketplace prices are high.

Second, we have evidence of Amazon's preference for a hybrid marketplace instead of either a pure 1P model or a pure 3P model. Study 2b has clear evidence of a mix of 1P and 3P wins at equal prices. We find evidence from our other studies that also support Amazon's preference for a hybrid marketplace. For example, if Amazon intended to self-preference, it would not need to respond to large 3P price increases because the Buy Box would ensure that the 1P seller would win. Instead, in Study 1, we see evidence of 1P price cuts after a big rise in 3P prices, and 3P sellers subsequently cut their prices (likely to improve their Buy Box win chances). That is, Amazon cares about 3P prices being too high, a finding that would be harder to reconcile with self-preferencing in the Buy Box. Similarly, if Amazon wanted to self-preference, there would be much less need or evidence for entry when 1P prices are high. We see from Study 3 that Amazon supports 3P entry. All these pieces of evidence show Amazon's preference for a hybrid marketplace, where there are opportunities for 3P sellers to profit. Recall that some fees (e.g., referral) are ad valorem, and some (like fulfillment) are per unit. The mix of these at various prices within a category and across categories is likely to influence Amazon's self-preferencing or 3P preferencing decisions in the Buy Box.

If Amazon's margins are possibly larger on 3P sales than on 1P sales, a reasonable question is why Amazon participates in 1P sales at all. ⁶ As previously discussed, Study 1 shows that 1P presence may help in lowering 3P prices. Additionally, Amazon provides a steadier marketplace presence compared with 3P sellers—even primary 3P sellers delist from time to time. A steady 1P presence preserves its marketplace reputation among consumers, especially Prime subscribers who expect inventory and high-quality service even when 3P sellers delist. We infer that Amazon's 1P presence serves the purpose of both price moderation and stability.

Discussion

Implications for Marketing Stakeholders

We explore the broader implications of our findings for several marketing stakeholders.

For consumers

The implications of Amazon's dual role for consumers are quite benign. The large number of Prime customers pay fees for better services (e.g., faster shipping, easy returns, high-quality customer service, streaming deals). We saw from Studies 1 and 2 that Amazon uses its 1P price to counter high 3P prices and the Buy Box to penalize low-quality, inconsistent 3P sellers, ensuring consumers get both low prices and quality sellers. Khan (2017) argues that low prices can harm consumers if used to deter entry and then subsequently increase prices. Study 3 mitigates these fears because when Amazon prices are high, there is increased entry.

For 3P sellers

Our results suggest that Amazon wants a hybrid marketplace in which both 1P and 3P sellers serve demand via Buy Box wins. Primary 3P sellers with competitive prices have a fair chance of securing the Buy Box. However, there are some concerns for 3P sellers that we do not explore in this work due to the lack of data on fees. As shown in our analysis, for equal 1P and 3P prices there is a mix of 1P self-preferencing and 3P preferencing. If fees prevent 3P sellers from increasing their prices, there may be considerably more self-preferencing than 3P preferencing. In such cases, to achieve low prices while maintaining profits, 3P sellers might succeed by pivoting toward volume-driven strategies to make up for possible lower per-unit 3P profits.

Our Buy Box random forest predictions (Study 2a) reveal that secondary or less reputed 3P sellers may not be able to compete directly with 1P and primary 3P sellers via the Buy Box. These secondary sellers might look for alternative ways to sell outside the Buy Box. One option is selling specialty or lower-ranked brands, as e-commerce marketplaces like Amazon benefit from long-tail items (Brynjolfsson, Hu, and Smith 2006). Specialty sellers might consider strategic advertising to maintain visibility and differentiate from mainstream brands, allowing them to sustain higher prices with lower volumes.

For brands

Our results have implications for brands in the marketplace. The only article in marketing that addresses this issue somewhat is Bei and Gielens (2023, p. 271), which compares brand performance on two platforms in China—one each of pure 1P and pure 3P—conjecturing that hybrid marketplaces like Amazon “may breed more channel conflict because the online platform becomes a direct competitor of its 3P sellers and vice versa” (see also Jap, Gibson, and Zmuda 2022). We know from our results that secondary 3P sellers have a low probability of winning the Buy Box, which limits adverse seller behavior. Brands also need not worry about high 3P pricing harming their brand, as Amazon offers 1P price cuts. However, Amazon's low-price strategy could increase conflict with other channels that price higher. As Amazon's retail share grows, brand managers should focus more on external conflict with other channels than on the marketplace conflict between 1P and 3P sellers given the low incidence of same-brand price cuts.

For antitrust regulators

Our results may alleviate antitrust concerns in several ways. First, we find no indication of algorithmic collusion leading to higher prices (e.g., Hansen, Misra, and Pai 2021). That is, Amazon's dual role does not weaken competition through collusion and higher prices. Second, Amazon's low-price goals are likely to benefit consumer welfare, and our analysis of 3P seller entry indicates well-functioning competition. The findings of mixed 1P self-preferencing and 3P preferencing have milder antitrust implications than current studies that find only evidence for self-preferencing.

However, as previously noted, 3P sellers may have to make significant investments (e.g., using FBA) to win the Buy Box. Antitrust studies should explore how different Amazon's profits from 1P and 3P sales are and whether these differences are driven by cost differences not tied to Amazon's marketplace power (e.g., Amazon not incurring inventory carrying costs in 1P sales) and/or if they are driven by Amazon's exercise of marketplace power.

For academics

Our key finding of Amazon actively seeking lower prices contrasts with existing literature suggesting that algorithmic pricing and tacit collusion typically lead to higher prices (Musolff 2022; Wang et al. 2023). Researchers might consider a shift of focus away from algorithmic collusion on this marketplace given that Amazon's business model is based on scale/volume and therefore low prices (Economist 2017). Additionally, as previously mentioned, concerns about channel conflict in dual role/hybrid marketplaces may be less severe than previously thought in the literature (e.g., Bei and Gielens 2023). This opens opportunities to explore new models to examine whether hybrid marketplaces strengthen or harm brands.

More investigation is warranted into the mix of self- and 3P preferencing. The current research on self-preferencing has not explored categories with enough price variance within and across categories, especially in the Buy Box. As previously noted, Amazon's private label share is small; the only definitive data suggest 1%–3% in 2019. ⁷ Therefore, Amazon self-preferencing its 1P private label sales (as found by Waldfogel [2024]) must be viewed in the context of its significantly larger business in branded products, its larger margins on 3P sales compared with 1P on branded products, and its private labels being in systematically lower-priced categories.

Transportability of Findings to Today's Marketplace

Several notable shifts have occurred since our data period. This raises the question: Can our findings be relevant in today's landscape? We scraped additional data from September 24, 2023, through January 31, 2024, as noted in the “Data and Descriptive Statistics” section, and conducted robustness checks. Despite the shorter period of this data collection (where we see limited 3P price activity), the price change model (Study 1) for Amazon (see Figure 7 for key summary) shows similar insights: Amazon reduces prices in response to high Buy Box prices and to external competitor prices, and its own price adjustments are short-lived. ⁸

The continuity of our findings depends on whether Amazon still prefers lower prices and a hybrid marketplace. In the absence of longer tracking of current data, we turn to recent business press and analyst reports. Thus, these comments should be seen as exploratory rather than definitive.

Business press reports and data indicate that these preferences likely continue. Consider first the evidence for the continuation of a preference for low prices. Data collected by business intelligence companies suggest that Amazon continues to have the lowest prices among rivals. ⁹ A recent antitrust ruling also suggests that Amazon's low-price goals continue. As the Wall Street Journal reports, “German regulators are concerned Amazon's pricing tools may breach competition law. Amazon's systems highlight competitively priced goods and can remove overpriced listings. Amazon's tools can downgrade listings so they don’t feature prominently if it detects unusually high prices for certain products. … ‘If Amazon is prevented from helping people find competitively priced offers, it will lead to a bad shopping experience for them, as we’d need to promote uncompetitive or even abusive pricing in our store,’ the spokesperson said” (Orru 2025).

Next, consider whether Amazon's preference for a hybrid marketplace might continue. An equity analyst report from Morgan Stanley calculates the estimated 2025 margin as 10.54%–15.67% on 1P sales and 52.59%–68.14% on 3P sales (see footnote 6 for the equivalent numbers during our data period). At the same time, the 3P share of the marketplace is rising (Statista 2025a). This margin difference and rising 3P share both suggest possible continued preference for a hybrid marketplace in today's time.

We turn next to industry developments, including the rise of algorithmic pricing, stronger competitors like Walmart, and increased antitrust scrutiny, to see how they might intersect with Amazon's low price and hybrid marketplace preference.

First, there is increasing use of algorithmic pricing across industries, which can lead to higher prices through algorithmic collusion (Musolff 2022; Wang et al. 2023). However, as previously discussed, Amazon's reported low prices are not consistent with this. The impact of algorithmic pricing on Amazon's hybrid marketplace preference, that is, 1P–3P mix, is unclear. Algorithmic pricing is more readily applicable by Amazon for its 1P sales but may not be so by 3P sellers, which might shift the Buy Box to 1P, changing the 1P–3P win ratios.

Another key development is the rise of the Walmart marketplace. Its e-commerce share grew from 4.4% (2017) to 10.6% (2024) and Amazon's from 36.4% to 39.7% (Statista 2025b). In the absence of any collusion, the rise of Walmart may increase 3P bargaining power and hence reduce Amazon's margins on 3P sales. However, rivalry with Walmart may also reduce 1P margins. These possible shifts in margins on 1P and 3P sales might also change Amazon's hybrid marketplace sales mix. Amazon has also faced growing antitrust scrutiny since our data period. If anything, more regulatory pressure may reinforce Amazon's low-price preference and reduce self-preferencing (Waldfogel 2024). However, antitrust pressure might lower Amazon margins on 3P sales if those fees fall, which may again alter the 1P–3P mix on its marketplace.

These developments suggest a possibly mixed impact on Amazon's low-price goals, though the business press reports otherwise; that is, Amazon is still the lowest-priced marketplace. Likewise, industry changes with the rise of Walmart and greater antitrust scrutiny may alter the 1P–3P sales mix on the marketplace. However, business data on continued greater 3P profits for Amazon (Morgan Stanley reports) and the rising share of 3P sellers (Statista 2025a) suggest that Amazon's hybrid marketplace preference will also continue. To that extent, our results may persist even in today's environment; this is an empirically testable finding.

Conclusion

Using a three-part analysis, we examine whether Amazon's dual role weakens competition in its marketplace. Through 1P pricing, the Buy Box, and seller entry, Amazon does not weaken competition—in fact, it likely strengthens price competition on its marketplace. Amazon prefers lower prices and a hybrid marketplace. These results have quite benign implications for consumers, (reputed and low-priced) 3P sellers, and antitrust. Our results on historical data might be relevant for today's data to varying degrees, as we see in the replication using validation data from a shorter window. Our estimation framework is portable to today's Amazon Marketplace for newer insights.

Further research is needed to test our model on more categories and current data. In particular, more estimates of hybrid self- and 3P preferencing are needed across a wider range of prices within and across categories. Rising revenues of sponsored advertising fees from 3P sellers might also lead to increased 3P preferencing. Several other aspects of competition on Amazon remain unexplored. Given the rise of Walmart's marketplace, it would be interesting to estimate Walmart's within-marketplace competition and compare it with Amazon's while accounting for their competition with one another. An additional research area is to study price adjustments on Amazon and Walmart in response to cost shocks like tariffs and examine whether the mix of 1P/3P sellers changes and the role of 1P sales in pricing shifts as a result. We see our research as providing an early study in marketing exploring hybrid marketplaces, with many more avenues for future research.

Supplemental Material

sj-pdf-1-jmx-10.1177_00222429261417677 - Supplemental material for Does Amazon's Dual Role Weaken Marketplace Competition?

Supplemental material, sj-pdf-1-jmx-10.1177_00222429261417677 for Does Amazon's Dual Role Weaken Marketplace Competition? by Sharmistha Sikdar, Vrinda Kadiyali and Giles Hooker in Journal of Marketing