Abstract: Disintermediation, where providers and customers bypass an intermediary to transact, plagues the 1.5 trillion-dollar platform economy. Despite platforms’ mitigation efforts, little is known regarding either the magnitude of disintermediation or efficacy of mitigation policies, largely due to unobservability of disintermediation. We tackle these two challenges by first detecting disintermediation and quantifying its magnitude. A generalizable geo-analytic method is developed to match non-reserved nights on Airbnb with overnight stays captured by mobile location data. The discovered disintermediation rate of 22.4% in the Austin sample potentially translates into $1.1 billion annual revenue loss to Airbnb. We next propose a theoretical framework to synthesize key levers of effective mitigation policies, and then causally assess the efficacy of three policies, Instant Bookable, Airbnb Plus, and Superhost, by leveraging providers’ policy adoptions as natural experiment and Difference-in-Differences (DID) model. Analyses of nearly 2 million online reservation records and .6 billion individual-level location records, supported by a series of robustness studies, show that Instant Bookable and Airbnb Plus respectively deter disintermediation by 0.09 days (30.2% of the disintermediation days) and 0.07 days (23.5%) per week. The effect is further amplified among providers with greater customer repatronage, longer on-platform tenure, or no preference for longer stays. Further mechanism studies offer strategic insights regarding how to assuage leakage on two-sided platforms.

Abstract: The impact of home-sharing platforms on communities where they operate has been mixed, yielding both benefits and costs. As a result, local regulators are making important policy decisions to regulate these platforms. To do so effectively, policymakers must understand both the economic and societal impacts of these platforms on different stakeholders and the effects of the regulatory actions taken thus far. One key dimension of their impacts is household financial wellbeing. This study empirically investigates the net effect of Airbnb’s entry on financial wellbeing, measured by the rate of financial delinquencies, while also accounting for local regulations. We find that the entry of Airbnb in a geography results in a reduction of mortgage and auto loan delinquencies by 3.21% and 4.84% respectively. However, the implementation of local regulations can offset these effects, particularly more stringent ones such as those that require the presence of a host. To further understand the observed effects, we conduct extensive heterogeneous analyses to study the impact of moderating factors, such as social capital and the homestead exemption, and to the borrowers’ risk profile. Our results show that the impact of platform entry is more pronounced in communities with higher delinquency costs, while the impact of regulation on delinquency rates is more significant in areas with higher risk. We also study borrower behavior in response to the additional income and find evidence of increased borrowing, mostly driven by high-income hosts. Looking next at renters, who do not directly participate in home-sharing platforms, we find evidence of negative spillover effects. Finally, given the varied demographics of different sharing economy platforms, we examine potential differences in their impacts. To highlight these differences, we compare Airbnb with ride-hailing and food delivery platforms, and find varying loan repayment behaviors, which have different regulatory implications.   

Abstract: Online deceptive counterfeiting has become a significant source of loss for buyers, authentic sellers, and the retail platform itself. This problem is exacerbated on Amazon, the focus of this study, due to the proliferation of third-party sellers, some of whom might be plying counterfeit products, while blending in with genuine sellers. We develop a novel approach to identify the intensity of counterfeiting in Amazon product listings, then build an empirical framework to quantify the impacts of counterfeiting on consumers, authentic sellers and the platform itself. Applying our methodology to two different product categories, a fashion experience good and a utilitarian search good, we confirm that average consumer utility is decreasing in counterfeiting intensity, more so for expensive and popular products. We further find a substitution effect between product listings with high versus low intensities of counterfeiting.  We leverage our structural parameter estimates to run counterfactual experiments, which suggest that the deployment of detection algorithms to signal counterfeiting intensity in product listings would align the interests of authentic sellers with the welfare of the platform. Overall, our analysis provides a robust empirical framework for identifying deceptive online counterfeiting and understanding its impact on the  stakeholders of an online retail platform. 

Abstract: The introduction of a new product to existing product lines typically gives rise to two opposing effects – market expansion (e.g., obtaining more market share from competitors or creating new demand) and cannibalization (e.g., competition with other products offered by the same firm). However, in a multi-sided network, the effects of new product introduction may become more substantial or complex due to network effects. In this study, we consider the introduction of carpool rides on a ride-hailing platform and evaluate the impact on both riders and drivers. We use a unique and proprietary dataset with fine-grained trip-level information provided by a leading ride-hailing platform, and exploit the natural experiment of the introduction of carpool rides on this platform. Our results show that the introduction of carpooling services significantly increases overall spending and usage on ride-hailing services for riders. Additionally, it increases drivers’ earnings and the number of trips completed. We provide evidence of disparate impacts on drivers, with full-time drivers being crowded out by part-time drivers due to increased labor supply. Furthermore, riders, particularly those who were frequent users of ride-hailing services or relied on them when public transportation was unavailable, tend to spend more after adopting carpool services. We also examine the effects of carpooling on product and supplier cannibalization, productivity improvement, and changes in the optimal driver-rider ratio. We are among the first to study the impact of lower-tier products on a sharing economy platform and to provide insights for platform companies on how new product introduction affects their ecosystem and revenues. 

Shortlisted for ISS Cluster Best Paper 2023

Abstract: Dynamic-pricing algorithms facilitate frequent price adjustments to optimize sales. Yet, overly frequent price fluctuations may complicate consumers' purchase decisions. This paper empirically investigates how algorithm-driven price volatility influences the occupancy rates of more than 105,000 rental properties in New York City listed on Airbnb. Because properties on Airbnb can be booked up to 12 months in advance, we compile two price-volatility measures: a property's frequency of price changes across travel dates on a given booking date (i.e., volatility over travel dates) and a property's frequency of price changes across booking dates on a given travel date (i.e., volatility over temporal distances). For both measures, the occupancy rates increase from flat pricing to a certain degree of dynamic pricing. However, the occupancy rates start to decrease when prices become too volatile, controlling for the magnitudes of price-level variation. A series of mechanism checks suggest that price volatility across travel dates leads to quality concerns, whereas price volatility across temporal distances leads to fairness concerns. Our findings suggest optimal algorithmic pricing cannot be achieved without considering consumers' behavioral responses. 

Abstract: Does ratings inflation help or hurt consumers? Managers? To answer these questions, we build a dynamic pricing model, wherein consumers’ utility depends on product features including the consumer rating that evolves over time. Applying control theory, we compare the pricing strategies with and without ratings inflation. In addition, we empirically validate the model using market data on 10,265 Airbnb properties over five years from Austin, Texas. We find a counter-intuitive result that ratings inflation helps both managers and consumers. Under the dynamic pricing context, ratings inflation helps managers because they earn a higher long-term profit although consumer surplus decreases. However, the ratio of consumer surplus in dynamic pricing relative to that from static pricing increases as ratings increase. Furthermore, the price premium over the static pricing decreases as ratings increase. Consequently, consumers are least worse off when the rating approaches its ceiling value. Hence, ratings inflation helps consumers as well.  

Abstract: In this paper, we first research on the distance distribution of human mobility with single vehicle based on the driving data from a taxi company in South China. Different from conventional exponential distribution, we discover the mobility distance with taxi follows power-law distribution. Further, we proposed a model which may explain the mechanism for the power-law distribution: mobility distance is constrained by time and fare. Specifically, the relationship between fare and mobility distance follows piecewise function, and responds to individual sensitivity; the relationship between time and mobility distance follows significant logarithmic relationship. These two factors, especially the logarithmic relationship between time and mobility distance, may contribute to a power-law distribution instead of an exponential one. Finally, with a simulation model, we verify the significant power-law distribution of human mobility behavioral distance with a single vehicle, by supplementing factors of waiting time and fare.

Abstract: The abrupt closure of offices during the COVID-19 lockdown has allowed millions of workers to discover advantages of remote work and led to disruptive transition from conventional office settings to remote arrangements even as the pandemic recedes. However, for firms to determine whether or not to offer remote work as a permanent working arrangement, it is essential to understand how workers value and respond to remote working opportunities. In this paper, we construct a novel dataset that comprehensively details firm-level job postings and individual-level wage records and employ difference-in-differences estimation (i.e., logit and panel regression) to examine the impact of remote work on workers' turnover and salary outcomes. Our main analyses indicate that, after the lockdown has been lifted, the propensity of switching is positively influenced by provision of remote work. Furthermore, workers who switch to remote work accept lower wage than those in onsite workplaces.

Abstract: Disintermediation, where providers and customers transact bypassing an intermediary, has challenged the business model and dwindled profits of the multi-billion-dollar platform economy. Despite the platforms’ efforts to mitigate disintermediation, little is known regarding the extent of disintermediation or efficacy of the mitigation policies, largely due to unobservability of disintermediation. We tackle these challenges by designing a geo-analytic methodology to identify and quantify disintermediation by matching online Airbnb booking and offline granular mobile location data. We further leverage DiD with matching samples to causally examine the efficacy of four Airbnb policies; and finally propose a cost-and-benefit conceptual framework to interpret the findings and guide platform designs of mitigation policies. We find, for instance, a 5.4% of disintermediation in Austin, TX over Summer 2019; and Instant Bookable reduces disintermediation by 9%, with a stronger effect among the hosts without preference for long-term lease, with more repeated guests, and more hosting experience.