From Lemon Market to Managed Market: How Flagship Entry Reshapes Sellers’ Composition in the Online Market

Abstract

With the rapid development of e-commerce, ensuring product quality on online platforms has become increasingly important, especially in developing countries where market regulations are still underdeveloped. By treating different sellers offering the same brand’s products as an industry, this study examines the impact of flagship store entry on online product quality by constructing a multiple period difference-in-difference model and conducts detailed empirical tests using full-category and large-span data from Taobao. The empirical results demonstrate that flagship store entry not only prompts the exit of incumbent sellers and deters potential new entrants due to the competition effect, but also facilitates the exit of low-quality sellers while attracting high-quality sellers as a result of a consumer-learning effect. Consequently, the overall quality of the industry is improved, and this effect is more pronounced in high-priced and durable goods industries. The findings of this study have important implications for market structure design and online quality governance in online marketplaces.

1. Introduction

Online shopping has become a dominant mode of consumer purchasing worldwide, reshaping traditional retail landscapes and consumption patterns [1]. In the rapidly developing Chinese e-commerce market, Taobao.com (hereafter, Taobao, which means “hunting for treasures” in Chinese) is an online retail platform where small businesses or individual entrepreneurs trade with prospective buyers. Founded by Alibaba Group, Inc. in 2003, Taobao has become China’s largest e-commerce platform. By the end of 2024, Taobao had close to 9.4 billion registered users and more than 1 billion product listings on a given day. It processed roughly 3.56 trillion transactions in 2024, more than Amazon and eBay combined. However, this rapid expansion has not only created extensive opportunities for market participation but has also introduced multifaceted challenges related to platform governance.

As the platform continues to soar, for consumers, the exponential increase in product variety and seller numbers intensify the difficulty of identifying high-quality products, thereby exacerbating the issues of counterfeit and low-quality products on online platforms owing to information asymmetry. For example, Taobao has been identified as a notorious market by the administrative office of the United States Trade Representative since 2016 in response to significant rights holders’ concerns regarding large numbers of low-quality and counterfeit goods being openly sold on the platform. Therefore, ensuring product quality is a critical governance challenge for e-commerce platforms, particularly in developing countries where market regulations are still underdeveloped [2].

Recent studies have provided considerable evidence that online rating systems, third party certification mechanisms, and quality warranty can help consumers identify high-quality products and have a positive impact on the sales and price of high-quality products [3,4,5]. While prior research has addressed the “lemon market” problem in online markets based on signaling theory, little work has examined the impact of industry dynamics on overall quality in the online market [6].

Due to the reduction in entry and exit costs for firms, entry and exit dynamics are more prevalent in online markets compared to offline markets [7].However, existing studies have predominantly focused on firm entry and exit in offline markets [8,9]. Moreover, these studies commonly argue that new entrants suffer from a reputation disadvantage, limiting their ability to convey product quality information to consumers, thereby resulting in a weaker impact on the overall product quality in industry [10]. However, in the online market, new entrants may have operated offline for several years and thus do not necessarily suffer from a reputational disadvantage upon entry, and the entry of new firms into online markets is likely to produce different effects.

In order to develop the existing literature, and to advance the governance mechanisms of “lemon market” problems in the online market, we leverage the entry of flagship stores on Taobao as the research context, and investigate the impact of flagship entry on overall industry quality. In 2012, Alibaba Group strategically separated Tmall from Taobao, positioning Tmall as a dedicated platform for branded products by selectively inviting brand owners or authorized dealers of specific brands to operate flagship stores, thereby insuring the provision of high-quality products with guaranteed authenticity under those brands. Consequently, considering sellers offering the same brand within the e-commerce platform as a discrete industry, prior to flagship store entry, consumers could only purchase from third-party sellers whose product quality is uncertain, resulting in a pooling equilibrium. With the introduction of flagship stores, consumers are presented with a choice between third-party sellers supplying products of uncertain quality and flagship stores providing high-quality products, which reflects a semi-separating equilibrium in the market [11,12].

Moreover, unlike traditional offline markets, online markets offer consumers greater temporal and spatial flexibility. Beyond an expanded product selection, the buying and selling process in online markets is continuous, bidirectional, and occurs in real-time [13,14]. Therefore, consumers purchasing from an online marketplace can compare product descriptions between flagship stores and third-party sellers, as well as consult reviews from previous buyers, thereby forming a consumer-learning effect, which may help them to purchase high-quality goods and promote the exit of low-quality sellers and thereby improve the industry’s average product quality.

Based on this logic, by treating different sellers offering the same brand’s products as an industry, we first develop a theoretical framework to examine the mechanism through which the entry of flagship stores influences the industry’s average quality. Then, using full-category and large-span data from Taobao, we construct a multiple period difference-in-difference empirical model to test the hypotheses we proposed. Specifically, at the macro level, we analyze how flagship store entry affects industry entry and exit rates; and at the micro level, we investigate its impact on the survival risks of sellers with different quality types. Finally, we examine the impact of flagship store entry on the overall quality of the industry. Drawing on our empirical findings, we offer relevant policy and practical implications guidance.

This study makes several marginal contributions to the literature. First, it offers a novel perspective and empirical evidence on the governance of the “lemon market” on e-commerce platforms. Specifically, while the existing literature primarily focuses on evaluating the effectiveness of various signaling methods, insufficient attention has been paid to the impact of new entrants on industry dynamics and product quality in the online market. In this study, we examine how the entry of high reputation sellers affects the average product quality on online platforms. This analysis not only enriches the understanding of market structure design in the e-commerce context but also holds significant implications for enhancing product quality in online markets.

Second, this study integrates firm reputation into the research framework on new entrants and industry dynamics. The existing literature predominantly assumes that new entrants face reputational disadvantages, which constrain their competitive effects and their ability to influence industry product quality [10,15]. Drawing on the unique context flagship stores entering online markets, this study examines the impact of high-reputation entrants on industry dynamics, thereby addressing the gap in prior research concerning the consideration of entrant reputation [16,17].

Third, with the development of the internet, consumers’ information acquisition capabilities have significantly improved. This study innovatively incorporates consumer learning into the analysis of market competition. By examining the quality types of sellers entering and exiting the market, it indirectly explores the impact of consumer learning on industry dynamics. This approach addresses the gap in the existing literature, which primarily focuses on firms’ price and quantity competition while largely neglecting the role of consumer-learning behavior in market competition [18,19].

Finally, whereas prior empirical studies on product quality in e-commerce tended to rely on data from single product categories [3,20], our study uses a large sample of data with a large time span and multiple categories. This comprehensive approach enables a robust empirical examination of our hypotheses at both the industry and firm levels, thereby mitigating the potential biases associated with single-category analyses.

The remainder of this paper is organized as follows. Section 2 reviews the recent literature on the governance of product quality in the online market and flagship store entry. In Section 3, we develop hypotheses on how the entry of flagship stores affects sellers’ exit and entry decisions, as well as industry product quality. Section 4 and Section 5 present our methodology and regression results, respectively. Finally, Section 6 summarizes the main conclusions, policy implications, and limitations of this study.

2. Literature Review

2.1. Governance of the Lemon Market in E-Commerce Platforms

E-commerce platforms, which are characterized by their dispersed and anonymous transactions, inherently suffer from information asymmetry, which aggravates product quality uncertainty and undermines buyer confidence [21]. Therefore, previous studies have mainly focused on platform governance mechanisms based on signaling theory and most have demonstrated that signaling mechanisms, certification mechanisms, reputation mechanisms, and warranties can improve market transaction efficiency and social welfare [3,22,23].

However, due to difficulties in obtaining data, significant empirical research examining the strength and interaction of these mechanisms has only emerged with the advent of e-commerce platforms [24,25]. Regardless, the extensive empirical literature on the impact of these mechanisms on selling performance has yielded mixed findings [26]. For example, Hou (2007) [27] investigates the extent to which a seller’s own quality claims can be perceived as a quality signal by bidders on the eBay platform. The empirical results of his study indicate that a seller’s own quality claims have no effect on auction outcomes (e.g., odds of sale and price), regardless of the seller’s reputation. However, Lewis (2011) [28] used a dataset of over 90,000 eBay Motors listings and measured the amount of information provided by sellers. By conducting a series of hedonic regressions, they found that this information was significantly correlated with price, reserves, and sales rates. Similarly, Chen and Wu (2021) [29] investigate the role of reputation regimes in trade by exploring t-shirt exports in Alibaba and find that the substance of an exporter’s reputation has a significant effect on the margins of exports. Regarding certification mechanisms, Elfenbein et al. (2015) [20] empirically examine the impact of quality certification programs on consumer demand using quasi-experimental matching methods with camera product data on eBay. They find that quality certification programs can help consumers identify high-quality products or sellers on e-commerce platforms, and this positive effect varies with market- and seller-level attributes. Hui et al. (2025) [30] study a major redesign of eBay’s quality certification that removed most consumer reports from its criteria and added administrative data, and the empirical results show that the proportion of certified sellers becomes more homogenized across markets, and sales seem to become more concentrated toward large sellers.

Regarding the interaction between different signaling mechanisms, Hui et al. (2016) [31] explore the impact of reputation badges, buyer protection programs, and their interaction on eBay’s marketplace. They find that reputation badges have a positive signaling value but adding buyer protection reduces the premium for reputation badges and increases efficiency in the marketplace. These efficiency gains are achieved by reducing moral hazards through an increase in seller quality and by reducing adverse selection through a higher exit rate for low-quality sellers. Xue et al. (2023) [23] conduct an empirical study based on the sales data of hairy crabs from 2239 sample sites and find that platform certification and product reputation have complementary effects.

Overall, the majority of empirical studies concentrate on the eBay market [32,33], investigating the potential of specific governance mechanisms to enhance sellers’ sales performance or generate price premiums. These mechanisms are generally regarded as reflecting the reputation signals of high-quality sellers, with research focusing on their impact on the sales of individual sellers in an environment with information asymmetry. However, these studies neglect to consider the impact of consumer acquisition of product quality information through such signals on the overall distribution of seller quality types within an industry.

Furthermore, studies focusing on Taobao often assume a market structure in which all sellers offer products of uncertain quality [34,35]. This assumption overlooks the fact that following the introduction of flagship stores, the Taobao platform evolved into a semi-separating equilibrium market consisting of flagship stores that sell high-quality products and other sellers offering products of uncertain quality. Additionally, most previous empirical research is limited by a narrow product scope, relatively short data periods, and small sample sizes [36,37]. These limitations largely explain the inconsistencies in the findings of previous empirical studies.

2.2. Product Quality in the E-Commerce Market

A key problem in the online markets of developing countries is the lack of reliable provision of high-quality goods and services [38]. Enhanced transparency encourages sellers to maintain high-quality products, thereby improving market efficiency [39]. Therefore, previous studies have mainly focused on the mechanism design of online platforms when analyzing sellers’ product quality choices. For example, Winfree and McCluskey (2020) [40] explore the optimal fee structure under a dual-reputation model in which consumers have a quality expectation based on the reputation of the platform and sellers. They find that fees depending on prices exacerbate poor quality incentives for sellers. Yoon et al. (2021) [41] find that signals embedded in a project description on a platform and the design of a recommendation system can help buyers find suitable sellers by mitigating perceived uncertainty and its implied risk. Zhao et al. (2023) [42] examine the quality and pricing decisions of two competing sellers in the presence of online customer review mechanisms in online markets. By developing two-stage game-theoretical models and comparing equilibria, they find that the existence of online customer reviews tends to encourage sellers to increase quality and charge low prices in the early stages, and decrease quality and raise prices in the later stages. Yang et al. (2024) [43] evaluate the role of advertising in providing information to a platform regarding new product quality to solve the cold-start problem. They find that using ad propensity information to rank new products benefits both the platform and consumers.

A few studies have analyzed product quality on online platforms in terms of sellers’ self-selection when entering online markets. Chen et al. (2015) [44] study a firm’s choice between online and physical markets with respect to its product quality. By constructing a flexible framework, they find that a firm’s quality choices in the two markets are contingent on the convexity or concavity of its marginal costs. When further considering the transparency policies and competition in online markets, Chen et al. (2017) [45] find that an entrant firm with product quality lower than that of an offline incumbent may choose the physical market, whereas entrants with a quality higher than that of the incumbent may sell online. Wang et al. (2023) [46] investigate a quality differentiation strategy by establishing a game model in which a manufacturer sells two products with similar functions but different quality levels. They find that the impact of quality differentiation on equilibria depends on the distribution strategy, contract form, and level of product competition.

To sum up, extant studies primarily focus on sellers’ self-selection of product quality, analyzing the product quality choices of individual sellers without examining industry-wide average quality from a macro perspective. Furthermore, these studies do not investigate the impact of the entry and exit of sellers of different quality types on the overall product quality in the industry. Additionally, existing studies do not pay sufficient attention to the possible impact of changes in the market structure and the entry of new sellers on other sellers’ product quality.

2.3. Firm Entry and Average Product Quality in an Industry

Presently, research on the impact of new firm entry on average product quality in an industry primarily focuses on offline markets [47,48,49]. Most studies begin from the perspective of entry deterrence, positing that incumbent firms strengthen their competitive advantage by improving product quality to deter potential entrants [50]. However, these studies are generally based on oligopolistic market structures and assume that new entrants face reputational disadvantages due to their lack of transaction history [10,15,51]. In contrast, the online market is similar to a perfectly competitive market and flagship stores are endogenously associated with a high reputation. Therefore, the competition brought about by the entry of flagship stores and its impact on product quality can be more complex.

A few relevant studies on the entry of flagship stores have explored the circumstances under which brand owners should establish flagship stores in the e-commerce market from the perspective of supply chain management [52,53]. Kargin and Lamey (2025) [54] examine the extent to which flagship stores add value and focus on whether the influence of flagship stores on firm value is contingent on the utilitarian and hedonic nature of the products offered by the initiating firm. However, these studies focus on the physical market and do not adequately address the potential impact of flagship store entry on consumer perceptions of product quality.

The study most similar to our research is that by Jin et al. (2021) [55], who empirically examine the effects of flagship entry on consumers, platforms, and various sellers on platforms. This study finds that a flagship entry may benefit consumers by expanding the choice set, intensifying price competition within the entry brand, and improving consumers’ perceptions of parts of a platform. However, the study only focuses on the competition effects of the entry of flagship stores, while overlooking potential consumer-learning effects, which may have an impact on the distribution of different quality types of sellers in the market and the average product quality in the industry.

3. Theory and Hypotheses

It is well documented that new market entrants encounter difficulties acquiring reputation benefits. Reputation has been shown to have a beneficial effect on long-run players, enabling them to earn higher equilibrium payoffs than would be impossible in the absence of a reputation [56,57,58]. Regardless, consumers can generally distinguish flagship stores from other sellers because of their extensive online shopping experience [54]. Consequently, even in the absence of a strong reputation accumulated through numerous prior transactions, consumers hold elevated expectations regarding the quality of products sold by flagship stores and are willing to pay a premium for them [55,59]. Essentially, flagship stores embody a dual perception of high quality and strong brand reputation among consumers. Therefore, the entry of flagship stores constitutes direct competition with other sellers offering the same brand of products [55]. This dynamic is likely to increase the survival risk of incumbent sellers by shifting demand away, thereby facilitating their exit from the market. Using transaction data from 2010 to 2020, Jin et al. (2021) [55] provide empirical evidence that sales from sellers of the same brand significantly decline following the entry of flagship stores.

However, flagship entry in the online marketplace not only intensifies competitive pressure on third-party sellers but also changes the consumer perception of other sellers on the platform [55], which may offer a protective advantage to sellers providing high-quality products. The entry of a flagship store is likely to provide consumers with additional opportunities for learning. Because flagship stores are directly owned by the manufacturers of branded products, they possess more specialized information regarding production processes and product quality than other sellers offering the same brand [54]. On Taobao websites, flagship stores typically present comprehensive product quality information through images and videos, whereas other sellers generally disclose less detailed information [60]. In some cases, sellers replicate or provide appropriate product descriptions created by the flagship stores. Therefore, the entry of flagship stores not only involves direct participation in market competition but also indirectly establishes product quality standards for consumers. This dynamic encourages consumers to seek detailed quality information provided by flagship stores actively when making purchasing decisions regarding third-party sellers. Furthermore, consumers can purchase products directly from flagship stores to obtain first-hand experience, thereby accumulating valuable knowledge for future purchase decisions.

Customer reviews on online platforms facilitate extensive communication between previous and prospective customers regarding product quality [61], thereby reinforcing the consumer-learning effect associated with the entry of flagship stores. Tóth et al. (2022) [62] demonstrate that buyers’ signals serve not only as feedback to sellers but also as influential information for other potential buyers in their purchasing decisions. Furthermore, flagship stores typically achieve higher sales than third-party sellers of branded products, which is attributable to their high quality and strong reputation. Consequently, both the volume and content of consumer reviews for flagship stores significantly surpass those of other sellers. For example, the flagship store of the “Apple Store” on Taobao, which sells Apple mobile phones, has accumulated approximately 100,000 consumer reviews, whereas “Cool Union Digital,” another seller of Apple mobile phones, has garnered only a few hundred reviews.

Therefore, following the entry of a flagship store, existing customers can obtain product quality information from previous buyers on the flagship store, thereby significantly reducing the cost of searching for such information from other sellers. Consequently, high-quality sellers may gain a competitive advantage through the consumer-learning effect, whereas the market demand for sellers offering low-quality products is likely to decline. Therefore, the entry of flagship stores may exert a more pronounced influence on low-quality sellers’ exit from the market. Based on these considerations, we propose the following hypothesis.

Hypothesis 1a.

The entry of flagship stores causes more sellers selling low-quality products to exit the online market.

Incumbent firms with excellent reputations enjoy considerable advantages that undermine the competitiveness of new entrants [63]. Consumers’ skepticism regarding the reputation of new entrants raises entry barriers because consumers are unable to verify product quality prior to purchase, leading them to reject goods from new entrants [64,65]. Consequently, reputation is a pivotal factor in the formation of entry barriers [66]. In this context, the strong reputation of a flagship store may serve as an implicit entry barrier, thereby limiting the number of new market entrants [67,68].

Flagship stores also possess the financial resources necessary to support special events and promotional activities at the point of sale, thereby enhancing consumers’ brand experiences on e-commerce platforms [60]. This capability may indirectly reduce advertising and reputation building costs for new sellers and potentially attract new entrants. Regardless, the impact of reputation spillovers may vary according to the quality of potential entrants. High-quality sellers typically exhibit a greater propensity to invest in reputation building, whereas low-quality sellers often adopt low pricing strategies to capture market demand rapidly and achieve profitability [69]. Consequently, reputation spillovers from incumbent flagship stores are likely to be more attractive to high-quality sellers entering the market. Based on this logic, we propose the following hypothesis.

Hypothesis 1b.

The entry of flagship stores causes more sellers selling high-quality products to enter the online market.

The distribution structure of sellers of different quality types directly influences an industry’s average product quality [70]. In accordance with the postulations outlined in Hypotheses 1a and 1b, it can be deduced that following the entry of flagship stores, the industry’s average quality level is expected to rise because of the increased proportion of sellers offering high-quality products online. Based on this reasoning, we propose the following hypothesis.

Hypothesis 2a.

The entry of flagship stores improves the average product quality in the online market.

Furthermore, the impact of flagship store entry on overall industry quality improvement may depend on the product characteristics of the entering brand. Numerous studies on consumer search behavior have demonstrated that consumers’ willingness to seek information and learn about a product prior to purchase is directly related to its price [71,72]. This relationship arises because the primary goal of consumer search before making a purchase decision is twofold: to select a high-quality product in a market characterized by quality uncertainty and to compare prices among products of similar quality [73]. Consequently, consumer decisions to engage in further searches hinge on the marginal costs and benefits of that search [74].

When purchasing low-priced products, potential savings from searching are limited. Therefore, consumers’ willingness to acquire information about product quality and their incentives to learn are reduced [75]. This finding implies that flagship store entries involving lower-value products tend to generate weaker consumer-learning effects. Supporting this concept, Liu et al. (2016) [76] analyzed online consumer purchase decision data and found that consumers spend less time reading and analyzing review content when buying cheaper products.

Conversely, for high-priced products, consumers tend to be more cautious and less sensitive to price differences [77,78]. In such cases, consumers are more inclined to purchase from flagship stores that ensure high-quality products, thereby intensifying the competitive effects of flagship store entry and further reinforcing the resulting improvement in average industry quality. Additionally, even consumers with high price elasticity who are unwilling to pay a premium for flagship store products exhibit stronger incentives to search for quality information when buying from third-party sellers, enhancing their ability to identify low-quality products and, consequently, amplifying the quality-enhancing effect of flagship store entry. Based on this logic, we propose Hypothesis 2b.

Hypothesis 2b.

The higher the average price of branded products entering the online market, the greater the quality improvement effect on the industry due to the entry of flagship stores.

Furthermore, the durability of the products offered by a brand plays a pivotal role in enhancing product quality following the entry of flagship stores. Durable products characterized by a longer lifecycle typically require consumers to invest considerable time and effort in understanding product attributes and performance before purchase [79,80]. Consequently, low-quality sellers are more likely to exit the market after entering flagship stores. Conversely, for non-durable products, consumer sensitivity to quality is relatively low and the pre-purchase learning process is comparatively brief. Therefore, the quality-enhancing effect triggered by the entry of flagship stores tends to be weaker for nondurable products. Based on this rationale, we propose the following hypothesis.

Hypothesis 2c.

The higher the durability of branded products entering the online market, the greater the quality improvement effect on the industry due to the entry of flagship stores.

4. Methodology and Data

4.1. Sample Selection and Data

The primary data source for this study is seller-level information obtained from Taobao between April of 2016 and April of 2019. It should be noted that during this period, the majority of flagship stores were directly invited by the Alibaba Group to join the Taobao platform, rather than actively applying for participation. Furthermore, for the few flagship stores that have initiated applications to enter the online marketplace, the exact timing of their entry remains unknown to third-party sellers offering products of the same brand. Consequently, a flagship store’s entry into the online marketplace is exogenous for third-party Taobao sellers. For this reason, we employ the difference-in-differences regression model as the primary empirical testing method [81,82], where the experimental group consists of brands that had a flagship store entry between April of 2016 and April of 2019, and the control group consists of brands that still do not have a flagship store entry as of April of 2019.

Brands with flagship stores are likely to have strong brand recognition, whereas brands without flagship stores are likely to be unknown brands. To address this sample self-selection bias, we selected brands with popularity similar to those with flagship stores as the control group. The selection of brands for the control group was conducted as follows. ① During the sample period, the average number of sellers operating the brand exceeded 30, because a higher number of sellers managing a brand indicates greater brand recognition. This criterion helps mitigate the issue of self-selection bias related to flagship store entry and ensures that brands in the control group possess sufficient recognizability. ② The sales volume of the brand in the control group is situated between the minimum and maximum values of the sales volume of the brands in the experimental group of the same product category. This criterion ensures that the control and experimental brands have similar market sizes. ③ Remove brands whose name is not easy to identify because it can lead to confusion when we determine the primary brand of third-party sellers. After screening the brands of the products used in the experimental and control groups, the sellers who sold these brands were excluded.

Identifying the seller’s primary brand is a key aspect of this study. On the Taobao platform, third-party sellers offer products from multiple brands within the same product category [55,83]. We determined each seller’s daily primary brand by calculating the transaction volume and value of each brand. We further computed the monthly frequency of each seller’s identified primary brand to address the potential instability of the primary brand indicator. If a brand appeared more than 20 times within a month, it was considered to be the seller’s primary brand. Otherwise, the seller was regarded as having an indeterminate primary brand for the month and excluded from the sample. This approach helps filter out inactive sellers who have not conducted actual transactions.

After performing data cleaning procedures, we ultimately obtained an unbalanced panel dataset spanning from April of 2016 to April of 2019. Considering two months as the time unit, the panel consisted of 903,276 seller–time observations with 292,973 sellers spread across 19 time periods and 167 product brands distributed across 11 product categories. In this study, the product category includes makeup/perfume/beauty tools, outdoor/hiking/camping/travelling goods, men’s shoes, women’s shoes, beauty care, men’s clothing, women’s clothing/women’s boutique, health food/supplementary food, watches, leather and luggage/women’s bags/men’s bags as well as sports/yoga/fitness/ball games products, which is a pre-set categorization in Taobao. Within the panel, 71 brands were designated as the experimental group and 96 brands were assigned to the control group.

4.2. Estimation Model

We employed three empirical models to test the hypotheses proposed above. The first model investigates the influence of flagship store entry on industry dynamics, as well as the distribution of seller types among exiting sellers and new entrants, testing H1a and H1b at the macro level. The second model analyzes the effect of flagship store entry on sellers’ survival risk and the heterogeneity of this effect on sellers of different quality types, testing H1a and H1b at the micro level. In the third model, we establish three indicators to measure the average product quality in the industry and examine whether the entry of flagship stores can improve the average quality of the industry, and the heterogeneity in this quality improvement effect, testing H2a and H2b.

4.2.1. Flagship Entry and Industry Dynamics

Equations (1) and (2) are regression models of the impact of flagship store entry on the entry and exit of other sellers, where $i$ denotes brand and $t$ denotes time.

$${exit}_{it}=\alpha +{\beta }_1{did}_{it}+{\beta }_2{X}_{it}+{\gamma }_t+{\mu }_i+{\epsilon }_{it}$$

(1)

$${entry}_{it}=\alpha +{\beta }_1{did}_{it}+{\beta }_2{X}_{it}+{\gamma }_t+{\mu }_i+{\epsilon }_{it}$$

(2)

The explained variables ${exit}_{it}$ and ${entry}_{it}$ represent the exit and entry rates of the sellers in industry (In this study, we take different sellers selling the same brand’s products on the e-commerce platform as an industry. Therefore, the industry actually refers to a specific brand) $i$ in period $t$, respectively. According to research examining industry dynamics [84,85], a seller is an exiting seller only if they withdraw from the market until the end of the period and do not re-enter the market. A new entrant is defined as a seller who exists in period $t+1$ but does not exist until period $t$. Therefore, the exit rate is the ratio of the number of sellers exiting the market in period $t+1$ over the total number of sellers in period $t$, whereas the entry rate is the ratio of the number of sellers entering the market in period $t+1$ over the total number of sellers in period $t$.

The primary explanatory variable, denoted as ${did}_{it}$, captures flagship store entries. This variable serves as the interaction term between dummy variables $d_{it}$ and $t_{it}$. $d_{it}$ represents whether the brand belongs to the experimental group, taking on a value of one if the brand had a flagship store entry between April of 2016 and April of 2019, and a value of zero otherwise. If a brand has a flagship store entry during period $t$, $t_{it}$ will take on a value of one from the period of $t$ to April of 2019 and a value of zero outside this period. The regression coefficient ${\beta }_1$ indicates the impact of flagship store entry on industry dynamics. $X_{it}$ is a vector of control variables, ${\gamma }_t$ and ${\mu }_i$ represent time and industry effects, respectively, and ${\epsilon }_{it}$ is a random error term.

This study employs three control variables. First, we include the exit rate (Lexit) and entry rate (Lentry) for the last period of the industry because the entry and exit of sellers within the industry is a continuous process with inertia [86,87]. Incorporating lagged entry and exit rates helps mitigate the influence of the external industry environment on sellers’ entry and exit decisions, while also reducing potential bias arising from omitted unobserved variables. Second, considering different sellers selling the same brand’s products on an e-commerce platform as an industry, the average characteristics of sellers that affect the entry and exit rates of the online market were calculated. We include the average age of sellers in the industry (Age) [88], average number of sellers placed on consumer follow lists in the industry (Coll), and average amount of deposits paid by sellers in the industry (Deposit). Third, we included the industry sales growth rate (Growth) and number of sellers in the industry (Comp), which serve as measures of competition in the industry [89]. To control for the influence of the number of existing flagship stores of other brands in the same product category on the brand’s sellers, we introduce a variable measuring the competition effect, which is measured as the number of flagship stores in the same product category before the entry of the brand’s flagship store (Comp_joint).

Additionally, considering that the control variables may have lagged effects on industry exit and entry rates, all independent variables in the regressions were lagged by one period. Furthermore, except for the sales growth rate (Growth), all control variables are expressed in logarithmic form.

4.2.2. Flagship Entry and Seller Survival

To examine the impact of flagship store entry on the survival risk of other sellers accurately, we removed sellers established only after the entry of the flagship store of each brand. We employ the Cox proportional hazards model for estimation because it accommodates right-censored data [90], imposes no prior assumptions on the baseline hazard function, and does not require the specification of the distribution of firm survival times, thereby providing flexibility that is well suited to our study.

Solving the data censoring problem is important for survival analyses [91]. Considering the issue of left-censoring, we select newly established sellers after April of 2016 as the research sample to remove left-censored observations [92]. Considering the issue of right-censoring, when a seller had not withdrawn from the market by April of 2019, it was impossible to determine their age accurately due to the absence of precise exit timing information. Therefore, we assigned a value of zero to the right-censored sample data in the exit variable corresponding to a failure event [93]. Additionally, we excluded samples that first appeared in the database in April of 2019 because their survival times encompassed only a single period and were right-censored, which may potentially bias the estimation results.

Following the data cleaning procedures outlined above, an unbalanced panel dataset containing 133,895 seller–time observations and 52,240 sellers was obtained. The regression model is defined as follows:

$$\lambda \left(t,x\right)={\lambda }_0\left(t\right)\mathrm{e}\mathrm{x}\mathrm{p}\left({\beta }^{\prime }x\right)$$

(3)

where ${\lambda }_0\left(t\right)$ is the baseline conditional hazard rate, $\beta =({\beta }_1,{\beta }_2,\dots ,{\beta }_p)$ is the regression coefficient, and $\lambda \left(t,x\right)$ represents the hazard rate function of sellers. The covariates are control variables. Based on research on the survival of sellers in online platforms [94,95,96,97], we include the age of the seller (Age_s), number of sellers in consumers’ following lists (Coll_s), sales volume (Sale_s), percentage of positive reviews given by consumers to sellers. Generally speaking, sellers with high positive ratings are preferred by the platform, and consumers can also filter sellers based on positive ratings during the search process, so there is a direct relationship between positive ratings and seller survival. And since the effective positive rating is used in distinguishing the quality type of sellers, the control variable at this time is just positive rating, not the effective positive rating. (Goodrate_s), overall rate of increase in sales in the industry (Growth), competition in industry (Comp), and joint competition (Comp_joint), which are measured in the same manner as described above.

4.2.3. Flagship Entry and Industry Average Quality

Equation (4) represents the impact of flagship store entry on the average quality of the industry.

$${quality}_{it}=\alpha +{\beta }_1{did}_{it}+{\beta }_2{X}_{it}+{\gamma }_t+{\mu }_i+{\epsilon }_{it}$$

(4)

Here, ${quality}_{it}$ denotes the average quality of industry $i$ in period $t$, ${did}_{it}$ denotes the entry of flagship stores in industry $i$ in period $t$, and $X_{it}$ denotes the control variables affecting the overall quality of the brand in the online marketplace.

In this study, we employed three methods to represent the average product quality in the industry. First, we constructed effective positive ratings (${quality}_1$) to represent the average product quality in the industry. Most studies use positive ratings, which are equal to the ratio of the number of good reviews over the total number of reviews, to measure product quality in the online market. A higher positive rating indicates higher product quality. However, recent studies have found that consumers are often more willing to provide positive feedback and that the distribution of positive ratings in online marketplaces is highly skewed [98,99,100]. Therefore, following Nosko and Tadelis (2015), and Hui et al. (2018) [101,102], we propose effective positive ratings, which are defined as the number of positive reviews minus default positive feedback divided by the total number of reviews. Default positive feedback is formed by the automatic evaluation of the Taobao platform itself after a consumer has not evaluated products for more than 15 days after receiving those products.

Second, we use the average value of the description score (${quality}_2$) of a brand’s sellers to indicate the average quality level of the industry. Once a Taobao transaction is completed, consumers can ascertain whether the item they have purchased is consistent with the image and description of the product displayed by the seller. If the description is consistent, the seller’s score for this indicator is higher. Although this indicator does not directly measure product quality, the seller’s description of a product is generally positive, suggesting that it can indirectly measure product quality. We use the average value of the description score of a brand’s sellers to indicate the average quality of the brand.

Third, we use the percentage of gold medal sellers in the industry (${quality}_3$) as an indicator of average product quality [23]. In July of 2014, Taobao introduced the gold medal seller certification system to the online market, with Taobao acting as a certifier. Sellers are not allowed to apply for certification. Instead, Taobao uses big data to assess the overall service level of all sellers regularly. Consequently, the percentage of gold medal sellers among all sellers of each brand represents the average quality of the brand.

Furthermore, while the aforementioned three indicators are all indirect measures of product quality and the positive review rate is also subject to shortcomings such as the presence of fake reviews [100,103], the data available to the authors suggest that this is the most effective measurement method. The effective positive rating (${quality}_1$) and description score (${quality}_2$) are subjective evaluations made by consumers based on their shopping experience, which reflect their satisfaction with product quality [104]. In contrast, the gold medal seller certification (${quality}_3$) is an official evaluation given by the platform based on the seller’s performance over the past month [105]. Consequently, while each indicator individually may capture only one dimension of product quality, their combined use helps mitigate individual weaknesses and provides a more robust composite measure.

Finally, we control a set of seller-level variables with potential impacts on product quality based on the related literature [2,93,106]. We include the proportion of sellers of the brand who have signed the “seven day return policy” ($Seven$), proportion of sellers who have signed the “consumer protection agreement” ($Protect$), average age of sellers in the industry ($Age$), and average number of sellers placed on consumers’ following lists ($Coll$), as well as the degree of competition in the industry ($Comp$) and joint competition effect ($Copm\_joint$). These variables are measured in the same manner as described above and all control variables are taken as one period lagged, given that product quality improvement takes time.

4.2.4. Descriptive Statistics

Descriptive statistics are presented in

Table 1. Descriptive statistics.

Variables	Obs.	Mean	Std.dev	Min.	Max.
Industry level
Effective positive rating (quality1)	3128	0.6151	0.1208	0	1
Description score (quality2)	3128	4.8584	0.0822	4.1202	5
Percentage of Gold Sellers (quality3)	3128	0.0428	0.0439	0	0.5
Flagship store entry (Did)	3128	0.2919	0.4547	0	1
Previous entry rate (Lexit)	2786	0.3348	1.9518	0	31 a
Previous exit rate (Lentry)	2957	0.2622	0.1243	0	0.8333
Industry sales growth rate (Growth)	2942	1.8528	32.4459	−1	1163.704
Average age of sellers (Age) b	3128	10.4951	0.2918	8.0644	11.2588
Average seller collection (Coll)	3128	7.5284	1.1169	0	12.3332
Average deposit of sellers (Deposit)	3128	11.5116	0.5749	0	13.4765
Number of sellers of the brand (Comp)	3128	4.8518	1.2870	0 c	8.7055
Seven day policy (Seven)	3128	0.9582	0.0586	0.5962	1
Consumer protection agreements (Protect)	3128	0.7952	0.1499	0	1
Competition effect (Comp_joint)	3128	4.0424	0.7104	2.7726	5.2257
Seller level
Seller’s age (Age_s)	903,276	10.0515	1.0428	3.2406	11.8430
Seller collection (Coll_s)	862,173	4.6171	2.4170	0	16.0122
Seller sales (Sale_s)	829,448	11.5109	1.9240	0	15.5001
Seller’s positive rating (Goodrate_s)	903,276	0.9908	0.0336	0	1

Note: ^a The maximum value of the entry rate is greater than one, indicating that the number of new sellers entering the next period is much higher than the number of sellers in the current period. ^b Age, Coll, Deposit, Comp, and Comp_joint are descriptive statistics after taking the logarithm of the variable. ^c The minimum number of sellers for the brands in the table is fewer than 30 for two reasons. First, the number of sellers for a particular experimental group brand may be fewer than 30. Second, brands in the control group were selected based on brands with an average number of third-party sellers greater than 30 during the sample period. Therefore, the control-group brands did not have more than 30 third-party sellers in any period.

.

5. Results

5.1. Flagship Entry and Industry Dynamics (Hypotheses 1a and 1b at Macro Level)

We first test the impact of flagship store entry on sellers’ entry and exit decisions, as well as the distribution of seller types among exiting sellers and new entrants. The regression results are presented in

Table 2. Flagship store entry and industry dynamics (macro level).

Variables	(1)	(2)	(3)	(4)	(5)	(6)
	Exit	Exit_b	Exit_g	Entry	Entry_b	Entry_g
Did	−0.0062	0.0103 ***	−0.0115 **	−0.3540 ***	−0.0007	0.0102 **
	(−0.9261)	(3.1528)	(−2.0781)	(−5.6883)	(−0.0646)	(2.4563)
Lexit	0.0656 ***
	(3.4758)
Lexit_b		0.4911 ***
		(28.8547)
Lexit_g			0.0551 ***
			(2.9451)
Lentry				0.0045
				(0.2693)
Lentry_b					−0.0391 *
					(−1.8706)
Lentry_g						0.1474 ***
						(7.2341)
Age	−0.0389 ***	−0.0281 ***	−0.0119	−0.3905 ***	−0.0399 **	0.0005
	(−3.6045)	(−5.1157)	(−1.3205)	(−3.5168)	(−2.0609)	(0.0744)
Coll	−0.0125 ***	0.0055 ***	−0.0173 ***	0.0001 ***	0.0040	−0.0072 ***
	(−4.5036)	(3.8431)	(−7.3942)	(23.1781)	(0.8481)	(−3.9536)
Deposit	0.0082 **	0.0017	0.0023	−0.1248 *	−0.0347 *	−0.0020
	(2.5291)	(1.0713)	(0.8371)	(−1.8099)	(−1.6884)	(−0.9532)
Growth	0.0000	0.0000	−0.0000 **	0.0011	0.0000	0.0000
	(−1.1456)	(−0.1831)	(−2.0811)	(1.3516)	(−0.1720)	(1.0199)
Comp	0.0149 ***	0.0055 ***	0.0057 **	−0.0407 *	0.0019	−0.0230 ***
	(5.0773)	(3.6875)	(2.3282)	(−1.7407)	(0.3836)	(−12.1645)
Comp_joint	0.0075	−0.0017	−0.0311	−0.0645 *	0.0118	−0.0277 *
	(0.2925)	(−0.1327)	(−1.4599)	(−1.6700)	(0.2837)	(−1.6770)
Constant	0.5058 ***	0.3588 ***	0.4461 ***	6.1037 ***	0.8916 ***	0.4413 ***
	(3.3940)	(4.8467)	(3.5819)	(4.7591)	(2.5782)	(4.5115)
Industry effect	Yes	Yes	Yes	Yes	Yes	Yes
Time effect	Yes	Yes	Yes	Yes	Yes	Yes
R2	0.5004	0.4110	0.4437	0.3089	0.0138	0.3867
N	2605	2564	2564	2605	2559	2559

Note: Robust t-statistics are presented in parentheses. *** p < 0.01. ** p < 0.05. * p < 0.1.

.

The regression results (Table 2, Column (4)) suggest that the entry of flagship stores did inhibit the entry of new sellers, indicating that the entry of flagship stores does have competition effect, which is consistent with the findings of Jin et al. (2021) [55]. However, the regression results (Table 2, Column (1)) show that the entry of a flagship store does not significantly promote the exit of other sellers. This may be because sellers need some time to make decisions about exiting the market [8], and the impact of flagship stores entering the market on sellers’ exit decisions requires a longer lag effect. It is also possible that the entry of flagship stores encourages the exit of low-quality sellers while deterring the entry of high-quality sellers, which results in an insignificant impact of flagship store entry on the exit rate (Table 2, Columns (2)–(3)).

We investigated the distribution of quality types among both existing and new entrants. In this study, we used the effective positive rating (${quality}_1$) to measure the product quality of each seller. This study also uses two other indicators of product quality, quality₂ and quality₃, the regression results are similar to those in Table 2 and can be provided upon reasonable request. We categorize sellers with a product quality lower than 25% (including 25%) as low-quality sellers and sellers with a product quality higher than 75% (including 75%) as high-quality sellers. We calculate the ratio of low-quality to high-quality sellers among the existing sellers and consider this ratio as an explained variable. We then re-examine Equation (1), where Exit_b represents the percentage of low-quality sellers among the exiting sellers and Exit_g denotes the percentage of high-quality sellers among the exiting sellers. The regression results (Table 2, Columns (2) and (3)) demonstrate that the entry of flagship stores significantly increases the proportion of low-quality sellers among exiting sellers and reduces the proportion of high-quality sellers among existing sellers, verifying Hypothesis 1a.

Similarly, the proportion of high- and low-quality sellers among new entrants is calculated separately. Unlike the method used to distinguish seller quality type described above, we distinguish the quality types of new sellers according to product quality in the second period after a seller’s entry. Therefore, the new sellers considered have survived for at least two periods, avoiding the measurement error caused by only one period of seller survival. Entry_b and Entry_g denote the proportions of low- and high-quality sellers among incoming sellers, respectively. The regression results (Table 2, Columns (5) and (6)) confirm that the entry of flagship stores has a significantly positive effect on the increase in the proportion of high-quality entrants to the online market, verifying Hypothesis 1b.

Overall, the results of the above regression analysis indicate that although the competitive effects triggered by the entry of flagship stores significantly reduced the number of new entrants, the new entrants were mainly high-quality sellers, while the exiting sellers were mostly low-quality sellers. This finding indirectly supports the existence of consumer-learning effects. Consumers’ enhanced ability to discern product quality leads sellers to adopt different entry and exit strategies in response to competition.

5.2. Flagship Entry and Seller Survival (Hypotheses 1a and 1b at the Micro Level)

To gain preliminary insights into the impact of flagship store entry on the survival risk of other sellers, Figure 1 and Figure 2 present the Kaplan–Meier survival estimates (product-limit estimates) and hazard estimate curves for the two groups of sellers, respectively. The red curve in Figure 1 indicates that the brands already have flagship store entries, whereas the green curve indicates that the products sold by the sellers do not yet have flagship store entries. These figures illustrate that the survival risk of sellers following the entry of a flagship store is considerably higher than that of sellers without such entry.

Furthermore, we performed a regression analysis based on Equation (2). Given that different product categories may exhibit varying initial survival risks, we employed product category as a stratification criterion and conducted a stratified Cox regression analysis. The regression results are presented in

Table 3. Flagship store entry and seller survival risk.

	Seller Tyle		quality1		quality2		quality3
Variables		(1)	(2)	(3)	(4)	(5)	(6)
Did		0.2265 ***	0.1721 ***	0.2497 ***	0.2290 ***	0.2308 ***	0.2272 ***
		(12.7219)	(8.5237)	(13.9299)	(12.7782)	(12.9462)	(12.6895)
Did × bad			0.1002 *		0.0440
			(1.9584)		(1.1297)
Did × good				−0.2815 ***		−0.1671 ***
				(−8.4079)		(−5.1564)
Did × gold							−0.0313
							(−0.3462)
Age_s		−0.0683 ***	−0.0883 **	−0.0682 ***	−0.0682 ***	−0.0682 ***	−0.0682 ***
		(−44.4936)	(−49.860)	(−44.4698)	(−44.4826)	(−44.4887)	(−44.4855)
Sale_s		−0.0530 ***	−0.1137 ***	−0.0546 ***	−0.0532 ***	−0.0545 ***	−0.0528 ***
		(−12.1892)	(−22.5298)	(−12.5166)	(−12.2235)	(−12.4831)	(−12.0533)
Goodrate_s		−0.5252 ***	−0.5810 ***	−0.5586 ***	−0.5286 ***	−0.5504 ***	−0.5248 ***
		(−3.3530)	(−3.5112)	(−3.5859)	(−3.3753)	(−3.5305)	(−3.3498)
Coll_s		−0.0838 ***	−0.0684 ***	−0.0855 ***	−0.0838 ***	−0.0845 ***	−0.0837 ***
		(−18.0461)	(−13.4284)	(−18.3947)	(−18.0661)	(−18.2108)	(−18.0260)
Growth		−0.0023	−0.0344	−0.0017	−0.0023	−0.0023	−0.0023
		(−0.2378)	(−1.6348)	(−0.1759)	(−0.2338)	(−0.2371)	(−0.2370)
Comp		0.1192 ***	0.1300 ***	0.1223 ***	0.1195 ***	0.1188 ***	0.1192 ***
		(16.0311)	(15.6667)	(16.4274)	(16.0622)	(15.9889)	(16.0334)
Comp_joint		0.0056	0.0060	0.0018	0.0053	0.0065	0.0057
		(0.3908)	(0.3713)	(0.1289)	(0.3667)	(0.4556)	(0.3985)
Industry effect		Yes	Yes	Yes	Yes	Yes	Yes
Time effect		Yes	Yes	Yes	Yes	Yes	Yes
N		82,081	82,081	82,081	82,081	82,081	82,081

Note: Robust t-statistics are shown in parentheses. *** p < 0.01. ** p < 0.05. * p < 0.1.

, where the coefficients are regression coefficients, not risk ratios. A coefficient greater than zero indicates that the entry of flagship stores increases sellers’ survival risk.

The findings in Table 3 illustrate the impact of flagship store entry on the likelihood of seller survival. The coefficients of Did are significantly positive at the 1% level in all models, including all controls and fixed effects (Table 3, Columns (1) to (6)). This finding suggests that the entry of flagship stores increases the survival risk of incumbent sellers [107], offering micro-level evidence of a significant competitive effect induced by flagship store entry.

To examine the heterogeneity of this impact on sellers with varying product quality types, Columns (2) to (3), (4) to (5), and (6) in Table 3 present the regression results that distinguish sellers according to their effective positive rating (${quality}_1$), description score (${quality}_2$), and gold medal certification (${quality}_3$), respectively. The coefficient of Did ^× bad in Column (2) is significantly positive and the coefficient of the interaction term Did ^× good in Column (3) is significantly negative. This indicates that while the entry of flagship stores raises sellers’ survival risk, it disproportionately increases the risk for sellers offering low-quality products relative to those selling high-quality products, thereby confirming the presence of a consumer-learning effect. Columns (4) to (6) demonstrate comparable regression outcomes, although some of the coefficients are not statistically significant; however, the direction of the coefficients aligns with expectations, confirming Hypotheses 1a and 1b.

5.3. Flagship Entry and Industry Average Quality (Hypotheses 2a, 2b, and 2c)

To examine the impact of flagship store entry on an industry’s average product quality, a regression analysis was conducted using Equation (4). The three measures of seller quality outlined previously serve as the explained variables in this section. The coefficients of Did are significantly positive in all models, including all controls and fixed effects (

Table 4. Flagship store entry and industry average quality.

	Seller Type	quality1	quality2	quality3	quality1	quality1
Variables		(1)	(2)	(3)	(4)	(5)
Did		0.0232 ***	0.0063 **	0.0425 ***	0.0285 ***	0.0248 ***
		(3.8676)	(2.5117)	(4.0136)	(4.7159)	(4.1376)
Did × Price					3.7040 × 10−6 ***
					(5.4955)
Did × Dura						0.0539 ***
						(3.4223)
Seven.		−0.1003 **	(0.0134)	−0.1590 *	−0.1075 **	−0.1086 **
		(−2.1829)	(−0.6989)	(−1.9599)	(−2.3510)	(−2.3664)
Protect		0.2818 ***	0.0463 ***	0.4924 ***	0.2757 ***	0.2786 ***
		(16.0043)	(6.3078)	(15.8304)	(15.7069)	(15.8320)
Coll		0.0220 ***	0.0036 ***	0.0417 ***	0.0219 ***	0.0218 ***
		(8.9275)	(3.5366)	(9.5826)	(8.9394)	(8.8759)
Age		−0.1347 ***	−0.0158 ***	−0.2258 ***	−0.1315 ***	−0.1338 ***
		(−13.9594)	(−3.9190)	(−13.2482)	(−13.6745)	(−13.8918)
Comp		0.0035	0.0027 **	0.0137 ***	0.0033	0.0038
		(1.3213)	(2.4124)	(2.9300)	(1.2370)	(1.4167)
Comp_joint		0.0096	0.0015	0.0241	0.0366	0.0282
		(0.4194)	(0.1533)	(0.5946)	(1.5679)	(1.1999)
Constant		1.6682 ***	0.1363 **	2.8007 ***	1.5412 ***	1.5942 ***
		(11.8238)	(2.3167)	(11.2381)	(10.8355)	(11.1896)
Industry effect		Yes	Yes	Yes	Yes	Yes
Time effect		Yes	Yes	Yes	Yes	Yes
R2		0.2613	0.1234	0.2852	0.2693	0.2644
N		2957	2957	2957	2957	2957

Note: Robust t-statistics are shown in parentheses. *** p < 0.01. ** p < 0.05. * p < 0.1.

, Columns (1) to (3)). This trend indicates that the entry of a flagship store significantly improves the average quality of the brand’s products, confirming Hypothesis 2a. This reflects the combined effects of competition and consumer learning. The competition effect alters the exit decisions of incumbent sellers as well as the entry decisions of potential entrants, while the consumer-learning effect exerts heterogeneous impacts on sellers of different quality types. Consequently, the proportion of high-quality sellers in the market increases, leading to an improvement in the average quality level of the industry. The results for the other control variables are aligned with our expectations.

Furthermore, to test Hypotheses 2b and 2c, we construct Equations (5) and (6) based on Equation (4) to evaluate the moderating role of the price level of different brands and whether they are durable goods on the quality improvement effect of flagship store entry. For the reference price, we calculate the average price of all products in the brand’s flagship store. The durability attribute of the products is a dummy variable, with watches, bags, and sports goods regarded as durable goods and assigned a value of one, whereas healthcare products, skincare and cosmetics, clothing, and footwear are regarded as nondurable goods and assigned a value of zero.

$${quality}_{it}=\alpha +{\beta }_1{did}_{it}\ast {price}_i+{\beta }_2{X}_{it}+{\gamma }_t+{\mu }_i+{\epsilon }_{it}$$

(5)

$${quality}_{it}=\alpha +{\beta }_1{did}_{it}\ast {Dura}_i+{\beta }_2{X}_{it}+{\gamma }_t+{\mu }_i+{\epsilon }_{it}$$

(6)

The coefficients of the key explanatory variables in the regression results (The moderated effects regressions presented in this paper only report the regression results when the mean of effective positive ratings (quality₁) is the explained variable. Results of the other two explained variables are similar, and the results could be sent if required) (Table 4, Columns (4) and (5)) are all significantly positive, indicating that flagship store entry has a significant impact on a brand’s average quality. Furthermore, the higher the price level of the brand or durability level of the product, the stronger the quality improvement effect of flagship store entry, validating Hypotheses 2b and 2c.

5.4. Robustness Analysis

5.4.1. Robustness Tests for Hypotheses 1a and 1b at the Macro Level

The E-Commerce Law of the People’s Republic of China came into effect in January of 2019, strengthening the regulations for online sellers. This change may have led to some sellers exiting the market, potentially affecting the net effect of the entry of flagship stores. Therefore, we deleted the samples after January of 2019 and reran the regression of Equations (1) and (2). The results (

Table 5. Robustness tests for Hypotheses 1a and 1b at the macro level.

Variables	(1)	(2)	(3)	(4)	(5)	(6)
	Exit	Exit_b	Exit_g	Entry	Entry_b	Entry_g
Did	−0.0098	0.0088 ***	−0.0127 **	−0.3583 ***	−0.0008	0.0117 **
	(−1.4614)	(2.6992)	(−2.2606)	(−5.1080)	(−0.0769)	(2.5379)
Lexit	0.0921 ***
	(4.6590)
Lexit_b		0.4676 ***
		(26.0512)
Lexit_g			0.0931 ***
			(4.6870)
Lentry				0.0025
				(0.1389)
Lentry_b					−0.0610 ***
					(−2.7048)
Lentry_g						0.1296 ***
						(5.8265)
Age	−0.0214 *	−0.0442 ***	0.0167 *	−0.3795 ***	−0.0412 **	0.0049
	(−1.8508)	(−7.6264)	(1.7316)	(−3.0537)	(−2.0198)	(0.6214)
Coll	0.0021	0.0020	−0.0079 ***	0.0001 ***	0.0047	−0.0070 ***
	(0.7028)	(1.3248)	(−3.1398)	(23.3335)	(0.9364)	(−3.4170)
Depoist	0.0100 ***	0.0086 ***	0.0053 **	−0.0137	−0.0282	−0.0025
	(3.1565)	(5.5119)	(1.9895)	(−0.1828)	(−1.3059)	(−1.1549)
Growth	0.0000	−0.0001 ***	0.0000	0.0010	0.0000	0.0000
	(−0.9137)	(−4.5994)	(1.0072)	(1.1469)	(0.0976)	(0.6982)
Comp	0.0174 ***	0.0033 **	0.0057 **	−0.0455 *	0.0024	−0.0228 ***
	(5.4745)	(2.0550)	(2.1803)	(−1.7536)	(0.4393)	(−10.4510)
Comp_joint	0.0016	−0.0040	−0.0152	−0.0628	−0.0073	−0.0241
	(0.0579)	(−0.2914)	(−0.6453)	(−1.4469)	(−0.1593)	(−1.2437)
Constant	0.2095	0.4975 ***	−0.0140	4.7065 ***	0.9011 **	0.3857 ***
	(1.2812)	(6.2342)	(−0.1025)	(3.3121)	(2.4440)	(3.4351)
Industry effect	Yes	Yes	Yes	Yes	Yes	Yes
Time effect	Yes	Yes	Yes	Yes	Yes	Yes
R2	0.2789	0.4450	0.3169	0.3293	0.0167	0.2727
N	2272	2401	2401	2272	2236	2236

Note: Robust t-statistics are shown in parentheses. *** p < 0.01. ** p < 0.05. * p < 0.1.

, Columns (2) and (3)) indicate that the entry of flagship stores promote the exit of low-quality sellers while retaining high-quality sellers among existing sellers, which is consisten with the results in baseline model (Table 2, Columns (2) and (3)). Considering the impact of flagship store entry on potential incoming sellers, the results (Table 5, Columns (5) and (6)) indicate that flagship store entry encourages entry by high-quality sellers and discourages entry by low-quality sellers, which is consistent with the results of the baseline model (Table 2, Columns (5) and (6)). Overall, the results of the robustness analysis are consistent with those of the baseline regressions and verify the robustness of H1a and H1b at the macro level.

5.4.2. Robustness Tests of Hypotheses 1a and 1b at the Micro Level

In the regression model above, which examines the impact of flagship store entry on seller survival risk, we restrict the sample to sellers established before the entry of flagship stores. Consequently, this analysis does not account for the survival of sellers who entered the market after the establishment of flagship stores. While excluding these post-entry sellers allows for a more precise estimation of the impact of flagship stores on the survival of incumbent sellers, it also excludes the valuable information contained in the experience of post-entry sellers. Therefore, we incorporated these sellers and re-ran the regressions specified in Equation (3) as a robustness check, as shown in Columns (1) to (3) in

Table 6. Robustness test for Hypotheses 1a and 1b at the micro level.

Variables	Expanded Sample			Lag by One Period
	(1)	(2)	(3)	(4)	(5)	(6)
Did	0.2055 ***	0.2053 ***	0.2088 ***	0.1579 ***	0.1577 ***	0.1816 ***
	(12.7727)	(12.7559)	(12.9501)	(5.9923)	(5.9877)	(6.8092)
Did × bad		0.0705 *			0.0549
		(1.9098)			(0.8791)
Did × good			−0.2165 ***			−0.2343 ***
			(−7.5871)			(−4.7680)
Age	−0.0705 ***	−0.0704 ***	−0.0704 ***	−0.0866 ***	−0.0866 ***	−0.0865 ***
	(−50.2437)	(−50.2319)	(−50.2407)	(−36.6195)	(−36.6092)	(−36.6074)
Sale	−0.0476 ***	−0.0479 ***	−0.0489 ***	−0.1057 ***	−0.1059 ***	−0.1063 ***
	(−11.6744)	(−11.7378)	(−11.9576)	(−16.6059)	(−16.6250)	(−16.6879)
PP	−0.5600 ***	−0.5668 ***	−0.6042 ***	−0.6064 **	−0.6096 **	−0.6399 ***
	(−3.7244)	(−3.7713)	(−4.0569)	(−2.5332)	(−2.5479)	(−2.6867)
Coll	−0.0805 ***	−0.0806 ***	−0.0814 ***	−0.0783 ***	−0.0784 ***	−0.0796 ***
	(−18.4165)	(−18.4533)	(−18.6295)	(−11.6303)	(−11.6522)	(−11.8164)
Comp	0.1124 ***	0.1123 ***	0.1117 ***	0.1433 ***	0.1434 ***	0.1467 ***
	(16.1279)	(16.1164)	(16.0398)	(13.2833)	(13.2929)	(13.5642)
Growth	−0.1309	−0.1302	−0.1293	0.0155 ***	0.0155 ***	0.0154 ***
	(−1.0661)	(−1.0608)	(−1.0576)	(4.8156)	(4.8130)	(4.8106)
Comp_joint	−0.0039 *	−0.0039 *	−0.0018	−0.0108	−0.0106	−0.0125
	(−0.2899)	(−0.2921)	(−0.1330)	(−0.5150)	(−0.5094)	(−0.5974)
Industry effect	Yes	Yes	Yes	Yes	Yes	Yes
Time effect	Yes	Yes	Yes	Yes	Yes	Yes
N	94,027	94,027	94,027	50,654	50,654	50,654

Note: We distinguish between firm quality types based on effective positive ratings (${quality}_1$). The regression results with the description score (${quality}_2$) and gold medal seller (${quality}_3$) measures are similar and can be provided if required. Robust t-statistics are shown in parentheses. *** p < 0.01. ** p < 0.05. * p < 0.1.

.

Additionally, we consider the possible lagged effect of flagship store entry on firm survival risk. Based on the sample used in the baseline model, we lagged all explanatory variables by one period and regressed Equation (3), and the results are presented in Columns (4) to (6) in Table 6.

The regression results (Table 6, Columns (1) to (6)) are consistent with those presented in Table 3, validating Hypotheses H1a and H1b at the micro level.

5.4.3. Counterfactual Tests of the Difference-in-Differences Model

The fundamental assumption underlying the difference-in-differences approach is that in the absence of flagship store entry, the trends in the treatment and control groups follow parallel paths without systematic divergence over time. A potential concern in our sample is that flagship store openings are endogenous to the brand owners themselves. Notably, popular and well-established brands are more likely to establish flagship stores. Consequently, brand quality trends and seller survival probabilities may systematically differ between brands with and without flagship stores. To address this issue, we conduct two placebo tests to assess the robustness of our findings [108,109].

Counterfactual test 1: We introduce the variable treat2, which takes on a value of one if the brand has had a flagship store since October of 2017 and a value of zero otherwise. This approach allows us to investigate whether the average quality of these brands prior to October of 2017 differed significantly from that of the control group. According to the regression results (

Table 7. Counterfactual test of the difference-in-differences model.

Variables	quality1	quality2	quality3
	(1)	(2)	(3)
Counterfactual test 1
treat2	0.0857	0.0615	0.2202
	(1.0667)	(1.2613)	(1.4496)
Control variable	Yes	Yes	Yes
Industry effect	Yes	Yes	Yes
Time effect	Yes	Yes	Yes
R2	0.0639	0.0513	0.3185
N	948	948	948
Counterfactual test 2
Did *	−0.0051	0.0041	0.0007
	(−1.0667)	(1.6011)	(0.7389)
Control variable	Yes	Yes	Yes
Industry effect	Yes	Yes	Yes
Time effect	Yes	Yes	Yes
R2	0.1012	0.1777	0.0412
N	2957	2957	2957

Note: Robust t-statistics are shown in parentheses. *** p < 0.01. ** p < 0.05. * p < 0.1.

, Counterfactual Test 1), the coefficient of treat2 is not significant when any of the quality indicators are used to measure the seller. This result indicates that the difference-in-differences method was appropriate for this study.

Counterfactual Test 2: We construct a counterfactual event time by moving the flagship store entry up by two periods. We then construct the flagship store entry variable Did*, which indicates that the improvement in the average quality of the industry comes from the entry of the flagship store if the coefficient is not significantly positive. The regression results (Table 7, Counterfactual Test 2) reveal that the regression coefficients for Did* are insignificant, indicating that the difference-in-differences method is appropriate for this study.

5.4.4. Parallel-Trends Test for the Difference-in-Differences Model

If the parallel-trends hypothesis is satisfied, then a quality improvement effect will occur only after the flagship store has entered the market. In other words, there should be no significant difference in the average quality of the brands in the experimental and control groups before flagship stores enter the market. We employed the event study methodology to test the parallel-trends assumption [110] and set up the following regression model based on Equation (4).

$${quality}_{it}=\alpha +{\sum }_{j=-7}^{j=7}{\beta }_j{did}_{i,t-j}+{\beta }_2{X}_{it}+{\gamma }_t+{\mu }_i+{\epsilon }_{it}$$

(7)

In Equation (7), the explained and control variables are identical to those in Equation (4). However, the key explanatory variables ${did}_{i,t-j}$ are different. ${did}_{i,t-j}$ is a dummy variable and when the flagship store of the brand enters in period $t-j$, the variable takes on a value of one in period $t-j$ and the following periods. Otherwise, it takes on a value of zero. Because there are fewer data points before and after the flagship store entry for the seven periods considered, we uniformly regard the data outside the seven periods as seven periods. Therefore, ${\beta }_{-7}$ to ${\beta }_{-1}$ represent the quality difference between the experimental and control groups before flagship store entry, ${\beta }_0$ indicates flagship store entry in the current period, and ${\beta }_1$ to ${\beta }_7$ represent the quality difference between the experimental and control groups following flagship store entry.

In accordance with the findings of Wang (2013) [111], we consider the previous period, in which flagship store entry occurred, as the reference group for the model. If the coefficients ${\beta }_{-7}$ to ${\beta }_{-1}$ are statistically insignificant, then it can be concluded that the parallel-trends hypothesis of the difference-in-differences method is satisfied. Coefficients ${\beta }_1$ to ${\beta }_7$ can be considered to represent the dynamic trend of flagship store entry affecting the average quality of the industry.

Figure 3 presents the regression results with effective positive ratings (${quality}_1$), description score (${quality}_2$) and gold medal certification (${quality}_3$) as the explained variables. The horizontal and vertical axes represent the coefficients ${\beta }_{-7}$ to ${\beta }_{-1}$ and their 95% confidence intervals, respectively. From Figure 3, the coefficients of ${\beta }_{-7}$ to ${\beta }_{-1}$ are not statistically significantly different from zero, indicating that the quality of the products in the treatment and control groups has the same trend before the entry of the flagship store and the parallel-trends assumption is valid. Furthermore, some of the coefficients between ${\beta }_1$ and ${\beta }_7$ are statistically significantly positive, suggesting that flagship store entry has a positive impact on the average quality of the industry.

6. Conclusions, Limitations, and Future Research

6.1. Conclusions and Implications

This study examined the influence of flagship store entry on the overall quality of a brand and conducted detailed empirical tests using full-category and large-span data from Taobao. By treating different sellers offering the same brand’s products on the e-commerce platform as an industry, our findings demonstrate that flagship store entry not only prompts the exit of incumbent sellers and deters potential new entrants, which aligns with previous research on competitive effects in online markets (Jin et al., 2021) [55], but also facilitates the exit of low-quality sellers while attracting high-quality sellers as a result of consumer-learning effects. Consequently, the overall quality of the industry is enhanced. Specifically, at the macro level, after the entry of flagship stores, the proportion of existing sellers offering low-quality products, and the proportion of new entrants providing high-quality products increases. At the micro level, the entry of flagship stores increases incumbent sellers’ survival risk, and this effect is less pronounced for sellers offering superior-quality products. Furthermore, heterogeneity analysis revealed that the quality improvement effect is positively correlated with brand pricing and durability.

Based on these findings, this study offers several policy implications. First, e-commerce platforms should actively encourage brand owners to establish flagship stores online, as their “high quality and strong reputation” can improve overall industry quality through consumer-learning effects. Additionally, both online and offline markets should facilitate the entry of new sellers and support those offering high-quality products to build their reputations rapidly. Current regulatory approaches tend to rely on costly ex post punitive measures that target poor-quality goods. In contrast, promoting the market entry of high-quality producers and effectively signaling product quality to consumers can increase industry standards more efficiently. Policymakers may consider subsidies and tax incentives to motivate the participation of high-quality firms.

Second, platforms and government agencies should assist firms in expeditiously establishing credible reputations through mechanisms such as government procurement programs and market-based ratings, particularly for durable and high-priced products. This study underscores the fact that consumers’ limited knowledge of product quality is a key driver of low-quality purchases. Enhancing information disclosure channels can reduce consumer search costs and influence firms’ quality decisions, thereby improving the overall industry output quality. Furthermore, leveraging mobile internet technologies to facilitate frequent information exchange among firms and consumers could mitigate information asymmetry by strengthening the learning effects.

6.2. Limitations and Future Research

Although we employed various methods and robustness checks using comprehensive category coverage and extensive time-span data in our analysis of Taobao, several limitations remain that warrant further investigation.

First, the lack of product price data and consumer data constrains our ability to explore the potential mechanisms through which the entry of flagship stores affects third-party sellers’ pricing strategies, as well as to conduct an in-depth analysis of consumer-learning behavior. This represents a significant limitation of the present study. Future research should endeavor to collect more price data and consumer data, enabling a more thorough examination of how flagship store entry influences firms’ pricing decisions, thereby systematically clarifying its broader impacts.

Second, while our study incorporated the time effect into the empirical model to control for other time-varying factors, the platform itself continuously implements various measures such as counterfeiting detection and clearance to improve the quality of online sellers’ products. This changing environment makes it challenging to identify a platform’s behavior accurately, which can result in an overestimation of the quality improvement effect of the entry of flagship stores. Future studies should compile detailed information from Taobao’s official announcements to construct a precise timeline of platform governance strategies, thereby disentangling the influence of platform policies on seller entry, exit, and quality decisions.

Finally, the three product quality indicators utilized in this study serve as indirect proxies and may not fully capture true product quality information. Future research should incorporate richer and more comprehensive data to measure product quality levels directly and accurately in online markets.